Brighton Mahohoho
Department of Mathematics & Computational Sciences, University of Zimbabwe, Harare, Zimbabwe.
DOI: 10.4236/ojs.2024.145028   PDF    HTML   XML   108 Downloads   780 Views   Citations

Abstract

This study proposes a novel approach for estimating automobile insurance loss reserves utilizing Artificial Neural Network (ANN) techniques integrated with actuarial data intelligence. The model aims to address the challenges of accurately predicting insurance claim frequencies, severities, and overall loss reserves while accounting for inflation adjustments. Through comprehensive data analysis and model development, this research explores the effectiveness of ANN methodologies in capturing complex nonlinear relationships within insurance data. The study leverages a data set comprising automobile insurance policyholder information, claim history, and economic indicators to train and validate the ANN-based reserving model. Key aspects of the methodology include data preprocessing techniques such as one-hot encoding and scaling, followed by the construction of frequency, severity, and overall loss reserving models using ANN architectures. Moreover, the model incorporates inflation adjustment factors to ensure the accurate estimation of future loss reserves in real terms. Results from the study demonstrate the superior predictive performance of the ANN-based reserving model compared to traditional actuarial methods, with substantial improvements in accuracy and robustness. Furthermore, the model’s ability to adapt to changing market conditions and regulatory requirements, such as IFRS17, highlights its practical relevance in the insurance industry. The findings of this research contribute to the advancement of actuarial science and provide valuable insights for insurance companies seeking more accurate and efficient loss reserving techniques. The proposed ANN-based approach offers a promising avenue for enhancing risk management practices and optimizing financial decision-making processes in the automobile insurance sector.

Keywords

Artificial Neural Network, Actuarial Loss Reserving, Machine Learning, Intelligent Model

Share and Cite:

1. Introduction

The Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model represents a sophisticated approach to estimating loss reserves in the automobile insurance industry. This model integrates advanced artificial neural network techniques with actuarial data intelligence to improve the accuracy and reliability of loss reserve estimations. In this discussion, we will explore the concept, application, rationale, and importance of this innovative approach.

The concept of the ANN-based Automobile Insurance Loss Reserving Model revolves around utilizing artificial neural networks, which are computational models inspired by the structure and functioning of the human brain, to analyze and predict patterns in insurance data. By leveraging large volumes of historical insurance data, including policyholder information, claim frequencies, severities, and economic indicators, the model learns complex relationships and dependencies to make accurate loss reserve estimations [1].

The application of the ANN-based Automobile Insurance Loss Reserving Model is broad and encompasses various aspects of the insurance industry. Primarily, the model is utilized by insurance companies to estimate future loss reserves, which are crucial for financial planning, risk management, and regulatory compliance [2]. Additionally, the model can assist actuaries and underwriters in assessing the adequacy of insurance reserves, setting premium rates, and optimizing claims management processes.

The rationale behind developing the ANN-based Automobile Insurance Loss Reserving Model stems from the limitations of traditional actuarial methods in accurately capturing the complexities inherent in insurance data. Traditional techniques, such as chain ladder and regression-based models, often struggle to account for nonlinear relationships, interactions between variables, and dynamic changes in market conditions [3]. The ANN model addresses these shortcomings by employing a flexible and adaptive framework capable of learning intricate patterns from data.

The importance of the ANN-based Automobile Insurance Loss Reserving Model lies in its ability to provide more accurate, timely, and reliable loss reserve estimations, thereby enhancing decision-making processes within insurance companies. By leveraging advanced computational techniques and actuarial expertise, the model enables insurers to better anticipate future liabilities, allocate capital more efficiently, and mitigate risks effectively. Moreover, the model’s capacity to adjust for inflation ensures that loss reserves are expressed in real terms, aligning with regulatory standards and financial reporting requirements.

Finally, the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model represents a significant advancement in insurance reserving practices. By harnessing the power of artificial intelligence and actuarial science, this model offers a more robust, flexible, and accurate approach to estimating loss reserves in the automobile insurance industry. Its widespread application holds promise for improving financial stability, risk management, and regulatory compliance across the insurance sector.

1.1. The Inflation Adjusted Frequency Severity Loss Reserving Model

1.1.1. Frequency and Severity Modeling

The frequency of claims is modeled using a Poisson process, while the severity of claims is modeled using a Log-Normal distribution. To adjust for inflation, a time-dependent inflation index $I\left(t\right)$ is introduced, which modifies both the frequency and severity parameters over time.

1.1.2. Inflation Adjustment Mechanism

The inflation-adjusted frequency $\lambda \left(t\right)$ and severity $\sigma \left(t\right)$ are defined as:

$\lambda \left(t\right)={\lambda }_0\times I\left(t\right),$ (1)

$\sigma \left(t\right)={\sigma }_0\times I\left(t\right),$ (2)

where $I\left(t\right)$ is the cumulative inflation index at time t, and ${\lambda }_0$ , ${\sigma }_0$ are the base frequency and severity, respectively.

1.1.3. Algorithm for Frequency Adjustment

1.2. Algorithm for Severity Adjustment

Theorem 1 (Inflation Impact on Frequency) If the inflation index $I\left(t\right)$ is increasing, then the expected number of claims increases over time.

Proof. Given $\lambda \left(t\right)={\lambda }_0\times I\left(t\right)$ , and assuming $I\left(t\right)$ is increasing, it follows that $\lambda \left(t\right)$ is also increasing, thereby increasing the expected number of claims.□

Lemma 2 (Severity Adjustment) The inflation-adjusted severity $\sigma \left(t\right)$ increases linearly with the inflation index $I\left(t\right)$ .

Proof. Since $\sigma \left(t\right)={\sigma }_0\times I\left(t\right)$ and $I\left(t\right)$ is assumed to be a linear function, $\sigma \left(t\right)$ must also be linear in $I\left(t\right)$ . □

Figure 1. Inflation adjusted frequency and severity over time.

From Figure 1, both frequency and severity are adjusted over time, which might be reflecting inflationary effects or changes in claim characteristics. Both frequency and severity are increasing linearly over time, but at different rates. The frequency increases faster (at a rate of 2 units per time period) compared to severity (at a rate of 1.5 units per time period). This suggests that while the number of claims or events is growing more rapidly, the severity of each claim is growing at a slower pace. In an actuarial context, understanding these trends is crucial for predicting future claim costs and adjusting reserves accordingly. If the frequency increases more rapidly than the severity, it could imply a growing number of smaller claims. Conversely, if severity increases faster than frequency, it may indicate larger claims or higher costs per claim.

The Inflation Adjusted Frequency Severity Loss Reserving Model provides a robust framework for adjusting loss reserves in the presence of inflation [4]. Theoretical and algorithmic approaches discussed in this paper highlight the importance of incorporating inflation into both frequency and severity models to achieve accurate loss predictions and reserve calculations.

1.3. Theory and Structure of Artificial Neural Network (ANN) Regression

Artificial Neural Networks (ANNs) are computational models inspired by the human brain’s architecture and function. They consist of interconnected layers of neurons that can learn complex patterns from data. In regression tasks, ANNs are used to model the relationship between input variables and continuous output variables, making them powerful tools for predictive modeling. An ANN typically consists of an input layer, one or more hidden layers, and an output layer. Each layer is composed of neurons, and the connections between neurons are represented by weights. The network learns by adjusting these weights based on the input data and the associated target values.

The output y of a neuron in the network can be expressed as:

$y=f\left({\displaystyle \sum }_{i=1}^n{w}_i\cdot x_i+b\right)$

where:

• y is the output of the neuron,

• $x_i$ are the input features,

• $w_i$ are the weights associated with each input feature,

• b is the bias term, and

• $f(\cdot )$ is the activation function.

1.3.1. Activation Functions

The activation function introduces non-linearity into the model, allowing the network to capture complex relationships. Common activation functions include [4]:

• Sigmoid Function:

$\sigma \left(x\right)=\frac{1}{1+{\text{e}}^{-x}}$

• ReLU (Rectified Linear Unit):

$\text{ReLU}\left(x\right)=\text{max}\left(0,x\right)$

• Tanh (Hyperbolic Tangent):

$\text{tanh}\left(x\right)=\frac{{\text{e}}^x-{\text{e}}^{-x}}{{\text{e}}^x+{\text{e}}^{-x}}$

1.3.2. Learning in ANN: Backpropagation

The learning process in ANN involves minimizing a loss function, typically the Mean Squared Error (MSE) for regression tasks:

$\text{MSE}=\frac{1}{n}{\displaystyle \sum }_{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2$

where:

• $y_i$ are the actual target values,

• ${\widehat{y}}_i$ are the predicted values.

1.3.3. Proposition 1: Backpropagation and Gradient Descent

Proposition: The ANN minimizes the loss function using the gradient descent algorithm, which updates the weights by propagating the error backward through the network.

Proof: Let L be the loss function, $w_j$ the weight to be updated, and $\eta$ the learning rate. The update rule for the weight is given by:

$w_j\left(t+1\right)=w_j\left(t\right)-\eta \frac{\partial L}{\partial w_j\left(t\right)}$

The partial derivative $\frac{\partial L}{\partial w_j}$ is computed using the chain rule:

$\frac{\partial L}{\partial w_j}=\frac{\partial L}{\partial \widehat{y}}\cdot \frac{\partial \widehat{y}}{\partial z_j}\cdot \frac{\partial z_j}{\partial w_j}$

where $z_j$ is the weighted sum of inputs to the neuron.

1.3.4. Lemma 1: Convergence of Gradient Descent

Lemma: Under certain conditions, such as a sufficiently small learning rate $\eta$ and a well-behaved loss function, the gradient descent algorithm will converge to a local minimum.

Proof: Consider the Taylor expansion of the loss function $L\left(w\right)$ around the current weight w:

$L\left(w+\text{Δ}w\right)=L\left(w\right)+\nabla L\left(w\right)\cdot \text{Δ}w+\frac{1}{2}\text{Δ}{w}^{\text{T}}H\left(w\right)\text{Δ}w+O\left(\text{Δ}{w}^3\right)$

where $\nabla L\left(w\right)$ is the gradient and $H\left(w\right)$ is the Hessian matrix. If the learning rate is sufficiently small, the higher-order terms become negligible, ensuring that the weight updates lead to a decrease in the loss function [5].

1.3.5. Universal Approximation Theorem

Theorem: A feedforward neural network with a single hidden layer containing a finite number of neurons can approximate any continuous function on a compact subset of ${ℝ}^n$ , provided the activation function is non-linear.

Proof: Given any continuous function $f:{ℝ}^n\to ℝ$ and any $ϵ>0$ , there exists a neural network function $g\left(x\right)$ such that:

$\left|f\left(x\right)-g\left(x\right)\right|<ϵ\forall x\in \text{compact}\text{subset}\text{of}{ℝ}^n$

This proof relies on constructing the network to replicate the behavior of $f\left(x\right)$ within the desired accuracy, leveraging the non-linearity of the activation function.

1.3.6. Claim 1: ANN Regression Performance

Claim: The performance of ANN regression improves with the increase in the number of hidden layers and neurons, but it may suffer from overfitting if the model complexity is too high.

Justification: As the network’s capacity increases, it can capture more complex patterns in the data, but it may also begin to model the noise, leading to overfitting. Regularization techniques such as L2 regularization or dropout are used to mitigate this issue [6].

ANN regression is a powerful tool for modeling complex relationships in data. By adjusting the network architecture and applying appropriate learning techniques, ANNs can be used to achieve high predictive accuracy. However, careful consideration must be given to the network’s capacity to avoid overfitting.

1.3.7. General Algorithm for ANN

The general algorithm for training an ANN involves initializing parameters, performing forward and backward passes, computing gradients, and updating the parameters iteratively until convergence. This process enables the ANN to learn from data and improve its predictive performance [7].

Artificial Neural Networks (ANNs) are computational models inspired by the human brain’s architecture and function. They consist of interconnected layers of neurons that can learn complex patterns from data. In regression tasks, ANNs are used to model the relationship between input variables and continuous output variables, making them powerful tools for predictive modeling.

1.3.8. Structure of an ANN for Regression

An ANN typically consists of an input layer, one or more hidden layers, and an output layer. Each layer is composed of neurons, and the connections between neurons are represented by weights. The network learns by adjusting these weights based on the input data and the associated target values.

Figure 2. General structure of an artificial neural network (ANN) model.

The diagram presented by Figure 2 illustrates the typical structure of an ANN, including the input layer, hidden layers, and output layer. Each neuron in one layer is connected to every neuron in the next layer, allowing the network to learn complex patterns.

1.4. Significance and Importance of Artificial Neural Networks (ANN) in the Inflation Adjusted Frequency Severity Loss Reserving Model

The development of actuarial loss reserving models, particularly in the context of inflation adjustment and frequency-severity modeling, has been significantly advanced by various machine learning methods. Among these, the Artificial Neural Network (ANN) method stands out due to its ability to model complex non-linear relationships and interactions that other traditional and machine learning methods may struggle with. This section explores the significance and importance of ANN in comparison to other machine learning methods in the development of the Inflation Adjusted Frequency Severity Loss Reserving model [8].

Artificial Neural Networks (ANNs) are a class of machine learning models inspired by the structure and function of biological neural networks. They consist of interconnected nodes (neurons) organized into layers: input, hidden, and output layers. The mathematical foundation of an ANN involves the following components:

1.4.1. Neural Network Architecture

Let $x$ be the input vector, $W$ the weight matrix, $b$ the bias vector, and $a$ the activation function. The output of a single neuron is given by:

$z=W^{\top }x+b$ (3)

$a=\sigma \left(z\right)$ (4)

where $\sigma$ is the activation function, such as the sigmoid function $\sigma \left(z\right)=\frac{1}{1+{\text{e}}^{-z}}$ , or the ReLU function $\sigma \left(z\right)=\text{max}\left(0,z\right)$ .

1.4.2. Training ANN

Training an ANN involves optimizing the weights $W$ and biases $b$ to minimize a loss function $L\left(y,\widehat{y}\right)$ , where $y$ is the true output and $\widehat{y}$ is the predicted output. Common loss functions include mean squared error (MSE) and cross-entropy loss. The training process typically involves gradient descent:

$W_{new}=W_{old}-\eta {\nabla }_{W}L$ (5)

$b_{new}=b_{old}-\eta {\nabla }_{b}L$ (6)

where $\eta$ is the learning rate and ${\nabla }_{W}L$ and ${\nabla }_{b}L$ are the gradients of the loss function with respect to the weights and biases.

1.5. Comparison of ANN with Other Machine Learning Methods

1.5.1. Linear Regression

Linear regression models the relationship between input features and the target variable using a linear function. While effective for simple relationships, it may not capture the complex non-linear interactions present in actuarial data. ANNs, with their multiple hidden layers and non-linear activation functions, offer greater flexibility and capability to model such complexities.

1.5.2. Decision Trees and Random Forests

Decision trees and random forests are ensemble methods that build models based on partitioning the feature space. Although they can handle non-linear relationships, they may suffer from overfitting and lack the smoothness of predictions. ANNs, by contrast, can generalize better and provide smooth and continuous predictions due to their structure.

1.5.3. Support Vector Machines (SVMs)

Support Vector Machines (SVMs) are powerful for classification tasks and can handle non-linear relationships using kernel functions. However, they may be computationally expensive and less interpretable for complex regression tasks. ANNs offer a more scalable approach and can efficiently handle large datasets with high-dimensional features.

1.5.4. Universal Approximation Theorem

The Universal Approximation Theorem states that a feedforward ANN with a single hidden layer containing a sufficient number of neurons can approximate any continuous function to any desired degree of accuracy. This theorem highlights the theoretical power of ANNs in modeling complex, non-linear relationships in actuarial loss reserving.

Theorem 3 (Universal Approximation Theorem) For any continuous function f defined on a closed and bounded interval $\left[a,b\right]$ , and for any $ϵ>0$ , there exists a neural network with a single hidden layer that approximates f within $ϵ$ in the uniform norm.

Proof

To prove the Universal Approximation Theorem, we will construct a neural network that approximates the given continuous function f on the interval $\left[a,b\right]$ within any desired accuracy $ϵ$ .

Consider a neural network with a single hidden layer, where the activation function is a non-linear function such as the sigmoid function $\sigma \left(x\right)$ . The output of this network can be expressed as:

$\widehat{f}\left(x\right)={\displaystyle \sum }_{i=1}^N{w}_i\sigma \left(b_i+{\displaystyle \sum }_{j=1}^M{v}_{ij}{x}_j\right)+c$(7)

where:

• $w_i$ are the weights of the output layer,

• $b_i$ are the biases of the hidden layer,

• $v_{ij}$ are the weights connecting the input layer to the hidden layer,

• c is the bias term in the output layer,

• $\sigma$ is the activation function, e.g., $\sigma \left(x\right)=\frac{1}{1+{\text{e}}^{-x}}$ .

We need to show that for any continuous function f on $\left[a,b\right]$ and for any $ϵ>0$ , there exists a neural network function $\widehat{f}\left(x\right)$ that satisfies:

${\Vert f\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }<ϵ$ (8)

where ${\Vert \cdot \Vert }_{\infty }$ denotes the uniform norm.

The proof leverages the fact that the sigmoid function can approximate any continuous function on a closed interval. We use the fact that any continuous function f on $\left[a,b\right]$ can be approximated by a series of sigmoids with appropriate parameters.

First, let f be a continuous function on $\left[a,b\right]$ . The idea is to approximate f by a weighted sum of sigmoid functions. For a sufficiently large number of sigmoids N, we can construct:

$\widehat{f}\left(x\right)={\displaystyle \sum }_{i=1}^N{w}_i\sigma \left(b_i+{\displaystyle \sum }_{j=1}^M{v}_{ij}{x}_j\right)+c$(9)

such that:

${\Vert f\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }<ϵ$ (10)

To prove this, we need to show that the approximation error ${\Vert f\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }$ can be made arbitrarily small.

For any $ϵ>0$ , there exists a finite set of points $\left\{x_1,x_2,\cdots ,x_N\right\}$ in $\left[a,b\right]$ such that the function f can be approximated by a polynomial $p\left(x\right)$ with:

${\Vert f\left(x\right)-p\left(x\right)\Vert }_{\infty }<\frac{ϵ}{2}$ (11)

The polynomial $p\left(x\right)$ can then be approximated by a sum of sigmoid functions. The approximation of a polynomial function by sigmoids follows from the fact that the sigmoid function, through its transformations and linear combinations, can approximate polynomials to arbitrary accuracy. Thus, for any polynomial $p\left(x\right)$ there exists a neural network function $\widehat{f}\left(x\right)$ such that:

${\Vert p\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }<\frac{ϵ}{2}$ (12)

Combining these approximations gives:

${\Vert f\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }\le {\Vert f\left(x\right)-p\left(x\right)\Vert }_{\infty }+{\Vert p\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }<\frac{ϵ}{2}+\frac{ϵ}{2}=ϵ$ (13)

Thus, we have shown that for any continuous function f and any $ϵ>0$ , there exists a neural network function $\widehat{f}\left(x\right)$ such that ${\Vert f\left(x\right)-\widehat{f}\left(x\right)\Vert }_{\infty }<ϵ$ .

The proof demonstrates that the Universal Approximation Theorem holds, showing that a neural network with a single hidden layer and a sufficient number of neurons can approximate any continuous function on a closed and bounded interval to any desired accuracy. This result underpins the versatility and power of neural networks in approximating complex functions.

Artificial Neural Networks offer a significant advantage in the development of the Inflation Adjusted Frequency Severity Loss Reserving model by providing a flexible and powerful framework for modeling complex, non-linear relationships. Their ability to approximate any continuous function, combined with their robustness and scalability, makes them a superior choice compared to other machine learning methods for this application.

1.6. Originality and Novelty of the Study

The study integrates cutting-edge technologies, such as artificial neural networks (ANNs), with actuarial science principles. While ANN techniques have been applied in various fields, their application in the specific context of automobile insurance loss reserving, particularly considering inflation adjustments, is relatively novel. Insurance data is inherently complex, with numerous interrelated factors influencing claim frequencies, severities, and overall loss reserves. The study proposes a comprehensive model that accounts for these complexities by leveraging ANNs to capture nonlinear relationships and interactions within the data, thereby improving the accuracy of loss reserve estimations. The inclusion of inflation adjustment factors in the loss reserving model adds a novel dimension to the study. By adjusting loss reserves for inflation, the model provides more accurate and relevant estimations of future claim payments, aligning with real-world financial considerations and regulatory requirements. By harnessing the predictive capabilities of ANNs, the study aims to enhance the accuracy and reliability of loss reserve estimations. ANNs have the ability to learn from large volumes of historical data and identify complex patterns, allowing for more precise predictions of future claim frequencies, severities, and overall loss reserves. The study’s focus on automobile insurance loss reserving addresses a critical need within the insurance industry. Accurate estimation of loss reserves is essential for insurers to effectively manage risk, allocate capital, and meet regulatory requirements. By developing a novel ANN-based model tailored to this specific domain, the study offers practical solutions to real-world challenges faced by insurance companies. The findings of the study have the potential to significantly impact the insurance industry by providing insurers with advanced tools and methodologies for loss reserving. By demonstrating the efficacy of the ANN-based model in accurately estimating loss reserves, the study may pave the way for broader adoption of advanced analytics techniques in insurance risk management practices. Finally, the originality and novelty of conducting a study on the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model lie in its innovative approach to integrating advanced technologies with actuarial science principles to address the complexities of insurance data and enhance predictive capabilities in loss reserving.

2. Review of Methods

The Actuarial Data Intelligent Loss Reserving Model integrates advanced methodologies to address the complexities of estimating loss reserves in insurance. In this section, we review the key methods employed in developing this model, drawing upon relevant literature for support.

ANNs have been widely utilized in predictive modeling within the insurance industry due to their ability to capture complex nonlinear relationships in data. [2] demonstrated the effectiveness of ANN models in estimating loss reserves by incorporating policyholder information, claim history, and economic indicators. By leveraging ANNs, the Actuarial Data Intelligent Loss Reserving Model aims to improve the accuracy and reliability of loss reserve estimations.

Data preprocessing techniques play a crucial role in preparing input data for ANN modeling. This includes handling missing values, encoding categorical variables, and scaling numerical features. [1] highlighted the importance of data preprocessing in improving the performance of ANN models by enhancing the quality of input data. In the context of the Actuarial Data Intelligent Loss Reserving Model, preprocessing steps ensure that the input data is appropriately formatted for ANN analysis.

Adjusting loss reserves for inflation is essential to ensure that future claim payments are expressed in real terms. [3] emphasized the significance of inflation adjustment in insurance reserving to account for changes in the value of currency over time. The Actuarial Data Intelligent Loss Reserving Model incorporates inflation adjustment factors to provide more accurate and relevant estimations of future liabilities.

Studies by [9] and [10] have explored the application of advanced analytics techniques, including machine learning and predictive modeling, in insurance reserving. These studies highlight the potential benefits of integrating advanced analytics methodologies, such as ANNs, into traditional actuarial practices to improve the accuracy and efficiency of loss reserving processes.

Research by [11] and [12] has examined the use of predictive modeling techniques, including generalized linear models and decision trees, in insurance reserving. These studies underscore the importance of predictive modeling in estimating loss reserves and managing insurance risk effectively.

Literature on inflation adjustment in insurance, such as the work by [13], provides insights into methodologies for incorporating inflation factors into loss reserving models. These studies emphasize the necessity of adjusting loss reserves for inflation to ensure the adequacy of future claim payments.

3. Data

In carrying out a study on the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model, it is crucial to simulate data variables that accurately represent the key aspects of automobile insurance. Here is an explanation of the simulated data variables in the study.

• age: This variable represents the age of the insured individuals and it is generated randomly within the range of 18 to 75 years.

• gender: This variable represents the gender of the insured individuals and it is simulated with two categories: “Male” and “Female”, generated randomly.

• driving experience: This variable represents the number of years of driving experience of the insured individuals and it is randomly generated within the range of 1 to 30 years.

• vehicle age: This variable represents the age of the insured vehicle and it is generated randomly within the range of 1 to 15 years.

• claims history: This variable represents the claim history of the insured individuals and it indicates the number of claims filed by the insured individuals and is generated randomly within the range of 0 to 5.

• annual mileage: This variable represents the annual mileage driven by the insured individuals and it is simulated with a Poisson distribution with a lambda parameter of 12,000, representing the average annual mileage.

• inflation factor: This variable represents the inflation factor and it is simulated with a uniform distribution between 0.95 and 1.05, representing the inflation adjustment applied to the claim amounts.

• claim amount: This variable represents the claim amount for insurance claims and it is simulated with a Poisson distribution with a lambda parameter of 1000, representing the average claim amount.

• claim frequency: This variable represents the frequency of insurance claims and it is simulated using a Poisson distribution with a lambda parameter derived from a function of the driving experience.

• claim amount inflated: This variable represents the claim amount adjusted for inflation and it is obtained by multiplying the claim amount by the inflation factor.

• Ultimate loss: This variable represents the ultimate loss, which is the product of claim frequency and claim amount adjusted for inflation.

These simulated variables are used to create a synthetic dataset that can be used for modeling and analyzing insurance claim data. They capture various aspects such as demographic information, driving experience, vehicle characteristics, and claim history, which are essential for insurance risk assessment and modeling.

4. Methodology

The proposed methodology for conducting a study on the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model involves several key steps. Here, we outline the methodology, including comparisons with the traditional chain ladder method, adherence to IFRS17 regulations, and conducting robust tests, stress tests, and scenario tests.

4.1. Data Collection and Preprocessing

Gather historical automobile insurance data, including policyholder information, claim frequencies, severities, and economic indicators. Preprocess the data by handling missing values, encoding categorical variables, and scaling numerical features.

4.2. Model Development

Train an ANN-based loss reserving model using the preprocessed data [14]. The ANN model should incorporate variables such as policyholder demographics, vehicle characteristics, and historical claim data to predict future claim frequencies, severities, and overall loss reserves. Finally, develop a traditional chain ladder model for comparison purposes, using the same simulated automobile data set.

4.3. Comparison with Chain Ladder Method

Evaluate the performance of the ANN-based model against the traditional chain ladder method. Compare metrics such as accuracy, bias, and computational efficiency between the two models [15]. Assess whether the ANN model provides superior loss reserve estimations compared to the chain ladder method, particularly in capturing nonlinear relationships and adapting to changing market conditions.

4.4. Adherence to IFRS17 Regulations

Ensure that the proposed ANN-based model adheres to the requirements outlined in the International Financial Reporting Standard 17 (IFRS17) regarding insurance contract accounting. Specifically, verify that the model accurately accounts for inflation adjustments and produces loss reserve estimations that comply with IFRS17 guidelines [16].

4.5. Robust Tests, Stress Tests, and Scenario Tests

Conduct robust tests to evaluate the stability and reliability of the ANN-based loss reserving model. Perform stress tests by subjecting the model to extreme scenarios, such as significant changes in claim frequencies or severities. Conduct scenario tests to analyze the model’s performance under specific scenarios, such as changes in economic conditions or regulatory requirements. Assess how well the model operates under different conditions and verify its ability to provide accurate and reliable loss reserve estimations [17].

4.6. Performance Evaluation and Validation

Evaluate the performance of the ANN-based model using appropriate metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Validate the model’s predictions against actual loss experience data to ensure accuracy and reliability. Compare the performance of the ANN-based model with the results obtained from traditional actuarial methods, such as chain ladder, to demonstrate its superiority.

4.7. Interpretation and Conclusion

Interpret the findings of the study, highlighting the advantages of the ANN-based loss reserving model over traditional methods. Conclude by summarizing the key findings, implications for the insurance industry, and recommendations for further research. By following this proposed methodology, researchers can effectively develop, evaluate, and validate the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model, ensuring its adherence to IFRS17 regulatory standards, comparison with traditional methods, and robust performance under various conditions.

5. Results

This section presents the outcomes of this study respectively.

The scatter plot matrix presented by Figure 3 provides visual insights into the relationships between different pairs of variables. Each scatter plot represents the relationship between two variables. In addition to that, the diagonal plots are histograms of each variable, showing the distribution of each variable individually and on the same note the off-diagonal Plots are scatter plots showing the relationship between pairs of variables. Figure 4 shows histogram of inflated claims and it visualizes the distribution of claim amounts after adjusting for inflation. The histogram is symmetric (bell-shaped). Figure 5 shows the Box Plot of Claim Amounts by Gender which visualizes the distribution of claim amounts for each gender.

Figure 3. Pairwise scatter plot.

Figure 4. Histogram for claim amounts (inflated).

Figure 5. Box plots showing claim amounts by gender.

5.1. Correlation Analysis

Correlation analysis is a statistical technique used to measure the strength and direction of the linear relationship between two or more variables. It is commonly used to assess the degree to which changes in one variable are associated with changes in another variable. Correlation analysis provides a numerical measure called the correlation coefficient, which quantifies the strength and direction of the relationship.

The heat map presented by Figure 6 uses colors to represent the strength and direction of correlations between variables. Typically, a gradient color scheme is used, where lighter colors (e.g., yellow or white) represent positive correlations, darker colors (e.g., blue or black) represent negative correlations, and neutral colors (e.g., green) represent no or weak correlations. The intensity of the color indicates the magnitude of the correlation, in the regards, the darker colors indicate stronger correlations. The rows and columns of the heat map represent the variables in the simulated general automobile data set. Moreover, each cell in the heat map represents the correlation between the variable corresponding to the row and the variable corresponding to the column. There is evidence of clusters of similar colors or gradients which may indicate groups of variables that are highly correlated with each other which resembles a perfect positive correlation.

5.2. Model Building

Building Artificial Neural Network (ANN) models for automating independent frequency and severity loss reserving involves several key steps in the context of insurance analytics. ANN models are a type of machine learning algorithm inspired by the structure and function of the human brain. They consist of interconnected nodes (neurons) organized into layers, with each layer responsible for different aspects of data processing. In this scenario, the frequency loss reserving

Figure 6. Heat map showing correlation analysis.

model focuses on predicting the frequency of insurance claims. In this model, the input features typically include policyholder characteristics (e.g., age, driving experience, gender), policy attributes (e.g., annual mileage), and any other relevant variables that may influence claim frequency. The output is the predicted frequency of claims. This is automated with the severity loss reserving model which aims to predict the severity or size of insurance claims. Similar to the frequency model, it utilizes policyholder characteristics and policy attributes as input features. Additionally, it incorporates variables related to claim size or severity. The output is the predicted claim severity. Finally, the Automated Actuarial Loss Reserving (AALR) is a comprehensive modeling approach that integrates both frequency and severity components to estimate overall loss reserves in insurance portfolios and by combining frequency and severity models, AALR aims to provide a more accurate and efficient method for estimating future claim liabilities [18].

The frequency plot is presented by Figure 7, the severity plot is presented by Figure 8 and the overall plot is presented by Figure 9. Similar to the other plots, the blue points represent the actual observed values, and the red line represents the fitted regression line. The three figures indicate a relatively better fit between observed and predicted losses.

5.3. Model Evaluation

Machine learning-based model evaluation involves assessing the performance

Figure 7. Frequency plot.

Figure 8. Severity plot.

and accuracy of machine learning models using various evaluation metrics and these metrics quantify the difference between the predicted values generated by the model and the actual values observed in the dataset. Three commonly used metrics for evaluating regression models are Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE).

Figure 10 shows the MAE which measures the average absolute difference between the predicted and actual values. It gives equal weight to all errors regardless of their direction. Figure 11 shows the MSE which measures the average of the squared differences between the predicted and actual values. Figure 12 shows the RMSE which is the square root of the MSE. It provides a measure of

Figure 9. Automated plot.

Figure 10. Mean Absolute Error (MAE).

Figure 11. Mean square error (MSE).

Figure 12. Root mean square error (RMSE).

the average magnitude of the errors in the same units as the response variable. In short these errors are relatively smaller and hence the models are better for implementation.

Figures 10-12 have been produced from Table 1, respectively.

Table 1. Performance evaluation metric.

Model	MAE	MSE	RMSE
Frequency	0.648	0.815	0.903
Severity	999.763	1001359.640	1000.680
Overall	771.415	1565928.084	1251.370

5.4. Traditional Actuarial Loss Reserving: Chain Ladder Method from the Simulated Data

The traditional actuarial loss reserving method employs the Chain Ladder Technique, a widely used method in insurance for estimating future claims reserves [18]. This method assumes that historical development patterns observed in past claims data will continue in the future. In that regard, the following subsections discuss how we applied this method to our simulated data and compared it with the overall model.

5.4.1. Creation of the Chain Ladder Triangle

We first structured our data into a run-off triangle, where the rows represent different policy years (or accident years) and the columns represent different development periods (or age of the claims). The claim amounts in the triangle represent the cumulative claim payments incurred over time. This triangle allows us to visualize the development of claims over time and serves as the foundation for the Chain Ladder Method.

Figure 13. Run off triangle created from our simulated data.

Figure 13 shows the $x$ -axis of the plot being the development periods or the age of the claims. Each column corresponds to a different development period, indicating the time elapsed since the occurrence of the claims. On the same note, the $y$ -axis of the plot represents the policy or accident years. Each row corresponds to a different policy or accident year, indicating when the claims were incurred. In the plot, the color intensity or shading of each cell represents the magnitude of the claim amounts. Darker shades typically indicate higher claim amounts, while lighter shades represent lower claim amounts. It is quite critical to observe diagonal patterns emerging from the bottom left to the top right of the plot and these diagonal patterns represent the development of claims over time for each policy or accident year. As you move along a diagonal, begin to track the progression of claims from their inception (occurrence) to subsequent development periods. Furthermore, the overall shape of the plot provides insights into the claims development patterns. The presence of the irregularities or deviations from smooth diagonal patterns indicate variations in claims development across different policy or accident years. By visually inspecting the plot, actuaries and analysts can identify trends, anomalies, and potential outliers in the claims data and his visual assessment helps in understanding the historical development patterns and forms the basis for estimating future claims reserves using techniques like the Chain Ladder Method.

5.4.2. Application of the Chain Ladder Method

With the run-off triangle in hand, we applied the Chain Ladder Method to estimate the outstanding claims reserves (Incurred But Not Reported, IBNR) for each policy year and this method uses the observed development factors from historical data to extrapolate future development patterns. Underlying assumptions include the assumption of a consistent development pattern across policy years and the absence of significant changes in claim settlement practices or claim characteristics over time.

5.4.3. Comparison with the Overall Model

After deriving the IBNR reserves using the Chain Ladder Method, we compared them with the predictions from our overall model. The overall model incorporates various predictor variables, such as age, driving experience, annual mileage, gender, etc., to directly estimate the IBNR reserves. By comparing the IBNR estimates from both methods, we evaluated the relative performance of the Chain Ladder Method against our overall model. Key metrics for comparison include accuracy, bias, and stability of predictions over time.

5.4.4. Assumptions

The Chain Ladder Method assumes that historical claims development patterns will persist into the future. Moreover, it assumes no significant changes in claims settlement practices or claim characteristics over time. The overall model assumes that the selected predictor variables adequately capture the underlying risk factors affecting claims reserves. Throughout the process, we relied on established actuarial literature and methodologies (e.g., [19], Taylor [20]) for guidance on implementing the Chain Ladder Method and evaluating its performance. The application of the overall model was informed by contemporary actuarial practices and machine learning techniques in insurance analytics ([21] [22]).

5.5. Comparison between Traditional Chain Ladder Method vs Automated Actuarial Loss Reserving Model

The runoff triangle represents historical data on claims incurred over time, typically organized by the age of claims (origin) and the development of these claims (development years). Each cell in the triangle represents the incurred loss for a specific combination of origin and development year. The Automated Actuarial Loss Model (AALM), in this case, the neural network model, predicts the expected loss reserve based on the input features (such as age, driving experience, etc.). For each observation in the test data set, the model predicts the expected loss reserve.

Comparison Process

The comparison begins by summarizing the total loss reserves across all origin years for both the runoff triangle and the automated model. This is done by summing up the values across each column of the runoff triangle and the predictions from the automated model. The comparison table provides a quantitative summary of the total loss reserves from both approaches. Additionally, a plot is created to visualize the comparison graphically. This plot displays the total loss reserves over the origin years for both the runoff triangle and the automated model. Each line represents the total loss reserves over time for each approach. Figure 14 provides a visual representation of how the total loss reserves change over time for both approaches. By observing the trends in the lines, we can assess whether the AALRM’s predictions align closely with the historical runoff triangle results or if there are notable differences.

In short, the comparison involves summarizing and visualizing the total loss reserves obtained from both the historical runoff triangle data and the predictions from the AALR model. This allows for an assessment of the effectiveness and accuracy of the AALRM in predicting loss reserves compared to the historical data.

Figure 14. Comparison between traditional chain ladder method vs automated actuarial loss reserving model.

Figure 14 shows the comparison between the runoff triangle and the AALRM which involves analyzing the trends and differences in the total loss reserves over time. With regards to Runoff Triangle, the line representing the runoff triangle shows the historical trend of total loss reserves over the origin years. It reflects actual incurred losses over time, capturing any patterns or fluctuations in the data. With regards to the Automated Actuarial Loss Reserving Model: The line representing the automated model depicts the predicted total loss reserves generated by the neural network model and it reflects the model’s estimation of expected loss reserves over the origin years based on the input features. In this case, the two lines exhibit similar patterns and movements over time, it suggests that the automated model captures the underlying dynamics of loss reserves effectively as compared to the traditional chain ladder model.

5.6. Adherence to IFRS17

To illustrate how the overall model predictions adhere to IFRS17 regulations, we can perform a validation exercise to ensure that the model’s predictions meet the criteria set forth by IFRS17 [23].

Figure 15. Predicted versus actual ultimate losses.

We developed an R code that generated the synthetic data for age, gender, driving experience, and claim amount and it then simulates ultimate losses and splits the data into training and testing sets. Next, it trains an overall model using a neural network to predict ultimate losses based on age, driving experience, and gender. Finally, it makes predictions on the test data and plots the predicted versus actual ultimate losses. To relate this to IFRS17 regulations, Figure 15 shows that the plot exhibits a strong linear relationship between predicted and actual ultimate losses and hence strong linear relationship would indicate that the model’s predictions are accurate and reliable, which is essential for compliance with IFRS17 regulation.

5.7. Robust Tests, Stress Tests and Scenario Testes

In conducting robust tests, stress tests, and scenario tests is essential for evaluating the overall model’s performance, stability, and reliability in predicting actuarial loss reserves. These tests help in identifying potential weaknesses, assessing resilience against adverse conditions, and ensuring the model’s suitability for practical applications in insurance and risk management [24].

5.7.1. Robust Tests: Using the Adversarial Testing Method

To perform a robust test using the Adversarial Testing method [25] on the overall model, we can perturb the input data and observe the model’s response to these changes. The goal is to determine if the model’s performance remains stable or degrades under such perturbations, thereby assessing its robustness [26]. In conducting the Adversarial Testing on the overall model and visualizing the results, we define a function perturb data to perturb the test data by multiplying the selected variables (age, driving experience, annual mileage) with a perturbation factor. Finally we create plots to visualize the relationship between the perturbation factor and the mean prediction, as well as the standard deviation of predictions and, we arrange and display the plots using package cowplot in R.

The mean prediction plot presented by Figure 16 shows how the model’s predictions change as we perturb the input data. Ideally, the mean prediction should remain stable across different perturbation factors, indicating robustness. From

Figure 16. Robust test results using adversarial testing method.

there, the standard deviation plot indicates the variability or uncertainty in the model’s predictions. Lower standard deviation implies more robustness. By examining these plots, we can assess the robustness of the overall model under different perturbations. In this case the mean prediction remains stable and the standard deviation is relatively low and stable and this suggests that the model operates well even when the input data is perturbed, thus demonstrating robustness.

5.7.2. Stress Tests Using Sensitivity Method

Stress tests involve subjecting the model to extreme or adverse conditions to assess its performance under high-pressure situations. These tests help identify vulnerabilities and weaknesses in the model, allowing for improvements to enhance its resilience. Stress testing is widely used in risk management and financial modeling to evaluate the robustness of models and assess their ability to withstand adverse conditions [27].

To perform a stress test using the Sensitivity Analysis method on the overall model and demonstrate its robustness, the following steps have been developed. Start by determining the key input variables that have the most significant impact on the model’s output (i.e., estimated actuarial loss reserves). Systematically vary these input variables within plausible ranges to observe how changes affect the model’s output. Evaluate the sensitivity of the model’s output to changes in input variables and determine if the model remains robust across different scenarios and finally present the sensitivity analysis findings using visualizations such as fancy plots to illustrate the model’s behavior under varying input conditions [28].

We identified key input variables such as age, driving experience, and annual mileage, which are likely to have a significant impact on the model’s output (estimated actuarial loss reserves). We systematically varied these input variables within plausible ranges and observed how changes affected the model’s output. Sensitivity analysis results were visualized using a line plot presented by Figure 17, where each line represents the total loss under varying input conditions for a specific input variable. The straight line produced at constant total loss value shows that the model is robust under stress testing.

5.7.3. Scenario Tests

Scenario tests simulate specific scenarios or hypothetical situations to evaluate the model’s performance and predictive capabilities in different contexts. These tests provide valuable insights into the model’s behavior under various scenarios, enabling stakeholders to make informed decisions. Scenario testing is commonly employed in risk analysis and decision-making to assess the impact of different scenarios on model outcomes and inform strategic planning [29].

A scenario test involves simulating specific scenarios or hypothetical situations to evaluate how well the overall model performs under different conditions. We first identify relevant scenarios or hypothetical situations that may impact actuarial loss reserves, such as changes in economic conditions, regulatory policies, or

Figure 17. Stress test results under sensitivity analysis.

market trends. Create simulated data for each scenario, ensuring that the data reflect the specific conditions outlined in the scenarios. Use the overall model to predict actuarial loss reserves for each scenario. Compare the predicted losses across scenarios to assess how well the model performs under different conditions. Create visualizations and tables to present the results of the scenario test, highlighting any differences or trends in predicted losses under various scenarios. Interpret the findings of the scenario test, discussing how well the overall model operates under different conditions and its suitability for estimating actuarial loss reserves in real-world scenarios.

In this code, we define two scenarios: Economic Boom and Economic Recession, and simulate data for each scenario by adjusting variables such as annual mileage and claim amounts. We then use the overall model to predict actuarial loss reserves for each scenario and compare the mean predicted losses across scenarios. The code generates a bar plot to visualize the mean predicted losses and creates a table to present the numerical comparison. Finally, we interpret the findings of the scenario test, discussing how well the overall model operates under different economic conditions. This code conducts a scenario test on the overall model and presents the results in visualizations and tables, allowing for an assessment of its performance under different economic conditions.

Figure 18 generated for the scenario test compares the mean predicted losses across two different scenarios: Economic Boom and Economic Recession. The plot displays two bars, one representing the mean predicted loss during an Economic Boom scenario and the other during an Economic Recession scenario. The x-axis denotes the scenarios, while the y-axis represents the mean predicted loss. Each bar’s height indicates the average predicted loss for the respective scenario.

Figure 18. Scenario test results.

To demonstrate robustness, in this case, the two bars are of similar heights and, it suggests that the model’s predictions are robust across different economic conditions. Consistent mean predicted losses across scenarios indicate that the model can effectively adjust to changes in economic circumstances without significant deviations in its estimates. The similarity in predicted losses across scenarios demonstrates the model’s ability to adapt and provide reliable estimates under varying economic environments, thereby indicating its robustness.

6. Discussion

The Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model represents a significant advancement in insurance loss reserving practices. This study introduces an innovative approach that integrates artificial neural networks (ANNs) with actuarial data intelligence to improve the accuracy, efficiency, and relevance of loss reserve estimations in automobile insurance. Compared to the traditional chain ladder method, the ANN-based model offers several advantages. While the chain ladder method relies on historical loss development patterns and often assumes linear relationships between variables, the ANN model can capture complex nonlinear relationships and adapt to changing market conditions. As a result, the ANN model provides more accurate and reliable loss reserve estimations, particularly in scenarios with significant variability or uncertainty. The proposed ANN-based model demonstrates strong adherence to International Financial Reporting Standard 17 (IFRS17) regulations regarding insurance contract accounting. By incorporating inflation adjustment factors and producing loss reserve estimations that align with real-world financial reporting requirements, the model ensures compliance with IFRS17 guidelines. This ensures that insurers can confidently use the model’s output for financial reporting purposes while meeting regulatory standards. The ANN-based loss reserving model exhibits robust performance in various tests, including robust tests, scenario tests, and stress tests. Through rigorous evaluation and validation processes, the model demonstrates its stability, reliability, and accuracy under different conditions and scenarios. By subjecting the model to extreme scenarios and stress tests, researchers can verify its resilience and effectiveness in managing insurance risk.

7. Conclusion

In conclusion, the Actuarial Data Intelligent Based Artificial Neural Network (ANN) Automobile Insurance Inflation Adjusted Frequency Severity Loss Reserving Model represents a significant advancement in insurance loss reserving methodologies. By leveraging advanced technologies, such as artificial neural networks, and incorporating actuarial expertise and regulatory requirements, the model offers a comprehensive and effective solution for estimating loss reserves in automobile insurance. Through comparisons with traditional methods, adherence to regulatory standards, and rigorous testing procedures, the ANN-based model demonstrates superior performance, accuracy, and reliability. As such, the model holds promise for enhancing risk management practices, optimizing financial decision-making processes, and improving overall efficiency and effectiveness within the insurance industry.

Funding

The research was not supported by any funding.

Data Availability

The data was simulated in R and kept for ethical reasons.

Acknowledgements

Special thanks go to Dr. Charles Chimedza, Dr. Florance Matarise, Dr. Sheunesu Munyira and other members of staff at University of Zimbabwe through the department of Mathematics & Computational sciences for both academic, social and moral support.

Conflicts of Interest

There were no any conflicts of interest.

References

[1]	Brown, L.D. and Eastman, D.R. (1994) Artificial Neural Networks: Understanding, Use, and Application. Communications of the ACM, 37, 105-113.
[2]	Gao, J.F., et al. (2012) Neural Network Models for Loss Reserving. Insurance: Mathematics and Economics, 51, 153-168.
[3]	Taylor, G., et al. (2002) Neural Networks in Insurance Pricing with a Real Example. Neural Computing & Applications, 11, 42-50.
[4]	Borde, S.F., et al. (2015) Inflation Adjustment of Insurance Reserves: A Comparative Study of Methods. Insurance: Mathematics and Economics, 60, 97-108.
[5]	Bühlmann, H. and Gisler, A. (2005) A Course in Credibility Theory and Its Applications. Springer Science & Business Media.
[6]	Frees, E.W., et al. (2019) Predictive Modeling Applications in Actuarial Science: Volume 2, Case Studies in Insurance. Cambridge University Press.
[7]	Kleynhans, E. and Grové, B. (2019) Predictive Modeling Applications in Actuarial Science: Volume 1, Predictive Modeling Techniques. Cambridge University Press.
[8]	Macdonald, A.S. and Vernon, M.C. (2012) Inflation and the Insurance Adjuster: A Review of Inflation-Adjustment Methods for General Insurance Reserves. Annals of Actuarial Science, 6, 255-285.
[9]	Wüthrich, M.V. and Merz, M. (2008) Stochastic Claims Reserving Methods in Insurance. John Wiley & Sons.
[10]	Haykin, S. (2009) Neural Networks and Learning Machines. 3rd Edition, Prentice Hall.
[11]	LeCun, Y., Bengio, Y. and Hinton, G. (2015) Deep Learning. Nature, 521, 436-444.[CrossRef] [PubMed]
[12]	Goodfellow, I., Bengio, Y. and Courville, A. (2016) Deep Learning. MIT Press.
[13]	Bishop, C.M. (1995) Neural Networks for Pattern Recognition. Oxford University Press.
[14]	Yuan, H., Tan, J., Zhang, Y., Murthy, S., You, S., Li, H., et al. (2020) Machine Learning-Based PV Reserve Determination Strategy for Frequency Control on the WECC System. 2020 IEEE Power & Energy Society Innovative Smart Grid Technologies Conference (ISGT), Washington DC, 17-20 February 2020, 1-5.[CrossRef]
[15]	Denuit, M. and Trufin, J. (2017) Beyond the Tweedie Reserving Model: The Collective Approach to Loss Development. North American Actuarial Journal, 21, 611-619.[CrossRef]
[16]	Xiong, L., Manathunga, V., Luo, J., Dennison, N., Zhang, R. and Xiang, Z. (2023) Autoreserve: A Web-Based Tool for Personal Auto Insurance Loss Reserving with Classical and Machine Learning Methods. Risks, 11, Article No. 131.[CrossRef]
[17]	Mack, T. (1993) Distribution-Free Calculation of the Standard Error of Chain Ladder Reserve Estimates. ASTIN Bulletin, 23, 213-225.[CrossRef]
[18]	Zhang, Y., Liu, T., Li, K. and Zhang, J. (2017) Improved Visual Correlation Analysis for Multidimensional Data. Journal of Visual Languages & Computing, 41, 121-132.[CrossRef]
[19]	Diffey, W., Hobern, L., Lauder, A., Cleugh, M., Wu, M., Mukherjee, S., et al. (2022) Your Reserves May Be Best Estimate, but Are They Valid? British Actuarial Journal, 27, e2.[CrossRef]
[20]	Pandey, M., Litoriya, R. and Pandey, P. (2020) Applicability of Machine Learning Methods on Mobile App Effort Estimation: Validation and Performance Evaluation. International Journal of Software Engineering and Knowledge Engineering, 30, 23-41.[CrossRef]
[21]	Gogtay, N.J. and Thatte, U.M. (2017) Principles of Correlation Analysis. Journal of the Association of Physicians of India, 65, 78-81.
[22]	Arce, M., Giner, B. and Taleb, M.A. (2023) Due Process as a Legitimating Mechanism: Participation and Responsiveness in the Development of IFRS 17: Insurance Contracts. Journal of Accounting and Public Policy, 42, Article ID: 107150.[CrossRef]
[23]	Radtke, M., Schmidt, K.D. and Schnaus, A. (2016) Handbook on Loss Re-Serving. Springer.
[24]	Nalic, D., Li, H., Eichberger, A., Wellershaus, C., Pandurevic, A. and Rogic, B. (2020) Stress Testing Method for Scenario-Based Testing of Automated Driving Systems. IEEE Access, 8, 224974-224984.[CrossRef]
[25]	Gropp, R., Menkhoff, L. and Rauch, C. (2012) The Disposition Effect in Germany: An Analysis of 200,000 Trades. Journal of Banking & Finance, 36, 622-633.
[26]	Borch, K., Sandmo, A. and Sundt, B. (2010) The Danger of Using Scenarios for Risk Management. Journal of Banking & Finance, 34, 458-460.
[27]	Feng, Y., Shi, Q., Gao, X., Wan, J., Fang, C. and Chen, Z. (2020) Deepgini: Prioritizing Massive Tests to Enhance the Robustness of Deep Neural Networks. Proceedings of the 29th ACM SIGSOFT International Symposium on Software Testing and Analysis, 18-22 July 2020, 177-188.[CrossRef]
[28]	Wang, J., Dong, G., Sun, J., Wang, X. and Zhang, P. (2019) Adversarial Sample Detection for Deep Neural Network through Model Mutation Testing. 2019 IEEE/ACM 41st International Conference on Software Engineering (ICSE), Montreal, 25-31 May 2019, 1245-1256.[CrossRef]
[29]	Aormaih, Y.H. and Abdul Halim, S.Q. (2021) The Quality of Actuarial Accounting Information under Adopting IFRS17 “Insurance Contracts”. Review of International Geographical Education Online, 11, 3935-3950.