One of the most important requirements for this research was a detailed historical and inclusive data set, which was acquired from the National Emergency Management Agency (NEMA), National Oceanic and Atmospheric Administration (NOAA) and the National Climatic Data Centre (NCDC) [9] [10] . The data used in this study covers the period from 1999 until 2019. The data collection sub-task is the process of identifying, extracting, and integrating log data from the source systems into a single repository. However, preprocessing is required to reduce the size of the dataset and transform it into a sliding window representation. Feature selection, the process of identifying a set of features from the data to be used in machine learning, is only performed for initial training and evaluation of the model. Therefore, the flood data was collected from different sources such as the National Emergency Management Agency (NEMA) and other publications. The details of the different sources, data of event, event type and references are given in Table 1 .

Data transformation operations are used to convert the dataset into an appropriate structure to facilitate machine learning. However, data aggregation and feature selection are common data transformation techniques used to obtain a reduced representation of the dataset without impacting its predictive accuracy [18] . Data pre-processing is required to transform the data into a format usable by machine learning algorithms. The data sets collated were inspected for outliers and extreme values, missing data and redundant information via a bespoke MATLAB application known as a data cleaning tool. This tool removes all existing outliers and missing data and re-orders the data based on specific categories chosen for the implementation of the ML techniques and it converts the alphanumeric and alphabetic data to numeric data using one-hot encoding. The processed dataset is then divided into training and testing data sets. The training data set is used to develop the model whereas the testing data set is used to quantify the accuracy of the model built. A larger portion of data is separated for training and the remaining is used for testing and validation to ensure accuracy of the classification model built and software performance. Figure 2 shows an overview of the overall analytical process employed in this study. The raw data collected is fed to the MATLAB data cleaning tool for data cleaning, normalization, aggregation, and other pre-processing steps. The output data is divided into testing and training data and passed through the ML/data mining application, the patterns are extracted, and the model is built, followed by analysis to verify its quality.

This study focused on supervised ML to learn from historical data, find clustered data, and build a classification model for future events. This type of ML works particularly best when used in combination with historical data (results included). For this purpose, several data mining tools such as orange canvas have been deployed. The reason for using the two software is to test more ML techniques with various training and testing dataset sizes. This software is user-friendly and can be easily accessible. The data will be divided into two parts. The first will be used for training and generating the model, and the second will be used for testing and verification. Several models were developed using different ML techniques to be able to measure and compare their performance and accuracy and choose the best. These techniques included Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Naïve Bayes (NB), and Logistic Regression (LR). The class for the model in all cases was set as “event type”, which included flood, flood/rainstorm, and flood/Windstorm. The independent attributes in all models were: location (community), state, population affected, injuries direct, injuries indirect, death direct, death indirect, property damage and crop damage.

Orange data mining software was originally developed by scientists at the University of Ljubljana in 1997 using the Python, Cython, C++ and C programming languages. The software’s graphical environment and interfaces have been developed using the Python and Qt3 libraries [19] . It opens commonly used dataset extensions such as txt, basket, CSV, arff. or Excel spreadsheet format. The method allowed input of climatic data such as rainfall variables (rainfall amount, intensity, duration, magnitude), it may also involve relative humidity, percentage relative humidity among others. The data could be uploaded and processed in the Orange Canvas software. This enables accurate prediction of like flood hazard over a long run (see Table 2 ).

Table 2 shows the flood data and its behavior over the period of study, it describe flood beginning, end and disaster recorded across the study areas. It also shows event type and state affected with population estimate within the beginning of location and end. The result is robust enough and could be reliable.

The system has been trained with several different combinations; however, the final system uses one based on the selected attributes, which was an output of the classifier attribute evaluation from an ML tool. All ML models developed were validated using evaluation criteria, i.e., confusion matrix [20] , Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) [21] . These metrics are used for summarizing and assessing the quality of the ML model. A confusion matrix summarizes the classifier performance concerning the test data. It is a two-dimensional matrix, indexed in one dimension by the actual class of an object and in the other by the class that the classifier allocates, and the cells represent: true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN) identified in a classification. Multiple measures of accuracy are derived from the confusion matrix i.e., specificity (SP), sensitivity (SS), positive estimated value (PPV) and negative estimated value (NPV). These are calculated as follows [22] :

$\text{SP}=\frac{\text{TN}}{\text{TN}+\text{FP}}$(1)

$\text{SS}=\frac{\text{TP}}{\text{TP}+\text{FN}}$(2)

$\text{PPV}=\frac{\text{TP}}{\text{TP}+\text{FP}}$(3)

$\text{NPV}=\frac{\text{TN}}{\text{TN}+\text{FN}}$(4)

The MAE is the mean of the absolute value of the error per instance over all samples in the test data. Each estimation error is the difference between the true value and the estimated value for the sample. MAE is calculated as follows [21] :

$\text{MAE}=\frac{{{\displaystyle \sum }}_{i=1}^n\left|y_{\text{estimate},i}-y_{\text{actual},i}\right|}{n}$(5)

where, $y_{\text{actual}}$ is the true value for the test sample i and $y_{\text{estimate},i}$ is the estimated or predicted value for the test sample i and n is the number of test samples.

The RMSE of a model for test data is the square root of the mean of the squared estimation errors over all samples in the test data. The estimation error is the difference between the true value and the estimated value for a sample. RMSE is calculated as follows:

$\text{RMSE}=\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^n\left|y_{\text{estimate},i}-y_{\text{actual},i}\right|}{n}}$(6)

Equations (1)-(6) were employed to validate the model, this step is also known as the model evaluation.

Descriptive statistics are used to summarize and describe the features of a dataset, providing insights into its central tendency, variability, and distribution. These statistics include measures such as mean, median, mode, standard deviations, and minimum and maximum values. Table 3 presents the summary of the south-south flood historical dataset. It was observed that the maximum duration of the flood event is 192 days and a minimum of 2 days. Between the periods of 1999-2019, a total of 59 deaths were recorded.

In addition, a plot of the flood events which reflects the two classes “Flood”, and “Flood/Rainstorm” indicates that “Flood” occurred more than “Flood/Rainstorm” in terms of duration ( Figure 3 ).

The 59 cases in the Orange Software test data have the target feature “Event-Type,” including 39 flood cases and 20 cases of flood combined with rainstorms. To determine which machine learning technique performs best, five various types of techniques are tested and evaluated in Orange Canvas software. The methods NN, LR, RF, NB, and SVM are tested. Figures 6-10 , which show the model training and testing procedure implemented in Orange, provide an overview of the process. To create the classification models, the training data is first run through various classification techniques (NN, LR, RF, and NB). The models are then tested on the test data. The study revealed that RF and SVM outperformed all other methods in terms of the percentage of classifications on the test data. The evaluation results and confusion matrix for the various ML models based on the provided test set are displayed in Figures 6-10 . The NB model categorized 8 out of 39 as flood and 18 out of 20 as flood/Rainstorm based on the confusion matrix ( Figure 4 ).

In the RF model, it was classified 35 out of 39 instances as Flood, and 18 out of 20 as Flood/Rainstorm. The correctly classified instances in total were 59 (100%) ( Figure 5 ).

In the LR model, it was classified that 39 out of 59 instances as Flood and 20 out of 20 as Flood/Rainstorm are correct. The correctly classified instances in total were 59 (100%) ( Figure 6 ).

In the NN model, it was classified that 39 out of 39 instances as Flood and 20 out of 20 as Flood/Rainstorm are correct ( Figure 7 ).

In the SVM model it was classified 39 out of 39 instances as Flood, 19 out of 20 were Flood/Rainstorm, and the correctly classified instances in total were 59 (100%) ( Figure 8 ).

The findings indicated that the south-south zone, particularly south-south settlements, is prone to flooding and rainstorms. The flood events at the starting and finishing locations are depicted in Figure 9 . Despite this, the print maps’ beginning and ending locations for the flood event do not significantly differ from one another ( Figure 9 ).

The study area’s maximum population affected by a flood occurrence is 50,000, as seen in Figure 10 .

Figure 11 demonstrates that, in the south-south region, the range of deaths brought on by flood events is 0 to 5.

The comparison of evaluation metrics from models constructed with both software tools and different test data sets shows that NB outperforms all other strategies, followed by RF. Note that the created model can be used to estimate the number of flooding incidents. As a result, the machine learning approach utilized in this work can provide insight into the patterns and frequency of flooding episodes, as well as the impact on population and property damage projected over a given period. Table 4 summarizes the model performance characteristics for the various machine learning techniques utilized.

F1 is a simple metric that involves the overall recall and precision of the model, while AUC is the area under the ROC curve, which is determined at thresholds between the True Positive Rate and the False Positive Rate. According to Figure 12 , NB (precision = 0.856) and RF (precision = 0.903) had the most accurate classifications of the flood event.

This study uses thirty years’ worth of historical flood data—which is extremely sparse because NEMA, the national disaster management agency, does not provide access to or availability of its data—to identify the types of floods that are most likely to occur in the future. Using MATLAB, the data was filtered to eliminate outliers, fill in missing values, arrange the data, and more. The machine was trained using 59 filtered instances (spanning the years 1990 to 2020) as the output while the remaining 25% of cases were used for testing. It is well recognized that Nigeria’s south-south region is particularly susceptible to the effects of climate change because of its location, climate, vegetation, soils, economic structure, population density, energy needs, and agricultural practices. The location, size, and distinctive terrain of south-south Nigeria result in a range of climates, from the tropical hinterland climate to the tropical maritime climate, which is typified by the rainforest along the country’s southern and coastal regions.

The location, duration, and effects of flood disasters on property and human life vary widely. To determine the kind of flood and its effects, it is necessary to consider several variables, including the location, duration, and geographic coordinates of the affected area. Assessing the flood event’s intensity and impact can also be aided by knowing the number of deaths, property and crop damages, and direct and indirect injuries that resulted from it. Nonetheless, the International Flood Event Classification System (IFECS), which divides floods into three categories—minor, moderate, and major—can serve as a basis for classifying floods. The length of the flood event, the extent of the impacted area, and the depth of the inundation all play major roles in determining this classification.

The results, which are consistent with [23] study, showed that floods and flood/Rainstorms are frequent in the southern region, especially in south-south settlements. Flood incidents are shown in Figure 4 for the initial location (communities), and Figure 5 for the final location. As the print maps ( Figure 4 and Figure 5 ) show, there isn’t much of a difference in the flood event between the beginning and ending locations. It was discovered throughout the study period that the flood had a significant effect on the destruction of farms and residences in the northern region, but the impact on homes (destruction of livable houses) was bigger in the southern location. About 400,000 people were most affected in 2000, as Figure 6 demonstrates. Similar effects of flooding were also found by [24] [25] [26] . On the other hand, direct repercussions include harm and deaths brought on by the flood itself, such as hypothermia, drowning, and injury from falling objects. The wider repercussions of the flood on human life, such as the interruption of necessary services, the loss of a means of subsistence, and mental health problems, are referred to as indirect impacts. Damage to property and crops are additional crucial aspect to consider when assessing the flood’s effects. These losses may be direct—caused by the flood’s immediate effects on buildings and farmland—or indirect—resulting from the aftermath of the incident.

Even so, there have been several significant factors that have influenced the development and evaluation of flood disaster models over time. For example, enhanced data collection and storage capabilities have made it possible to provide more precise and detailed model inputs, which has led to better simulations of flood events. The employment of increasingly intricate and sophisticated models has been made possible by developments in computer technology, producing simulations that are more precise and in-depth. Thus, assessing the precision and dependability of flood disaster models has required contrasting model simulations with actual flood occurrences. The comparison of evaluation metrics from models built using software tools and different test data sets reveals that NB beats all other strategies, followed by RF. Note that the generated model can be used to estimate the number of flooding episodes. As a result, the machine learning approach used in this study can provide insight into the patterns and frequency of flooding events, as well as the expected impact on people and property damage over time. The findings of this study are consistent with the reports of [27] and [28] . Although Rajab et al. rely on historical climate information. However, [7] found that prediction using machine-learning algorithms is useful since it can use data from several sources and categorize and regress it into flood and non-flood categories. Although the authors utilized Non-linear (NARX) and Support Vector Machine (SVM) machine learning techniques, they did not specify the best algorithm.