In the food and beverage sector, revenue forecasting is an essential aspect of strategic decision-making, especially for small and medium-sized businesses like coffee shops. Numerous variables, such as the number of customers, average expenditure, marketing initiatives, and foot traffic surrounding the company’s location, the number of employees, and working hours can affect daily income. For forecasting and planning, many coffee shop managers depend on their gut feelings rather than data-driven insights, even if they have access to key operational information ( Ahmad et al., 2020a ; Fiori & Foroni, 2020 ). Coffee shop operators may increase overall profitability, optimize marketing expenditures, and manage resources more effectively with the aid of accurate revenue forecasting. The use of straightforward, interpretable prediction models is lacking in many corporate settings, nevertheless, as resources for intricate systems could be scarce ( Munoz & Vassilvitskii, 2017 ).

To fill this gap, this study models and forecasts daily revenue using linear regression based on important business parameters. I determine the association between each parameter and daily income via an exploratory data analysis that includes statistical summaries, box plots, and scatter plots with regression lines. The most important variables, “Number of Customers Per Day”, “Average Order Value”, “Marketing Spend Per Day”, and “Location Foot Traffic”, are chosen to train the linear regression model. The main objectives of the study are as follows:

As predicting revenue is very crucial for all small and large businesses, many researchers are working in this field. Some of them are as follows: Lin and colleagues ( Lin et al., 2022 ) proposed employing Generalized Additive Models (GAMs) and machine learning models based on artificial neural networks (ANNs) to estimate the revenues of hybrid hydropower and energy storage systems in a quick and precise manner. With validation errors often less than 5%, the novel method cuts calculation time from around three hours to just one to four minutes per battery configuration when compared to standard MILP models. The accuracy of ML models was consistently higher than that of GAMs. This approach promotes the wider use of energy storage in renewable systems and provides investors and plant owners with a useful and effective tool for assessing battery size alternatives.

Researchers ( Jian et al., 2020 ) investigated the prediction of product market revenue using neural networks backed by fuzzy logic and artificial intelligence algorithms. Future sales are predicted using neural networks, which are well-known for their resilience and capacity to represent intricate nonlinear interactions. With prediction errors kept to 4%, the study concludes that the neural network-based prediction model achieves excellent accuracy. Businesses may increase profitability by using this forecasting technique to better understand market trends and make well-informed decisions.

A shortcoming of traditional frequent itemset mining that ignores the differing importance or monetary worth of transactions among various consumers is addressed in this work ( Weng, 2017 ). In order to address this, the authors developed a frequency-monetary (FM) weighting approach that more accurately reflects client value by taking into account both transaction frequency and revenue contribution. For mining high-revenue frequent itemsets from FM-weighted transactions, they put out a unique approach. The efficiency of this strategy in revenue-focused customer research was demonstrated by experimental findings utilizing survey data, which revealed that the top-k itemsets found using this method more correctly forecasted future customer revenues than the usual methodology.

In contract-based service contexts, where it’s critical to estimate changes in client or service-level revenues over time, the revenue change prediction problem addressed ( Mahajan et al., 2020 ). Since limited resources, like consultants or salespeople, must be effectively distributed based on anticipated revenue fluctuations, accurate forecasting is essential. By framing revenue change forecast as a classification issue, the authors present a unique paradigm. They present a system that optimizes prediction precision while reducing the loss of overall accuracy, acknowledging the problem of class imbalance in such datasets. The technique outperforms conventional classifiers and provides useful insights for resource prioritization. It is proven using real-world data from a top global cloud services provider.

The use of social media data, particularly YouTube trailer reviews, to forecast box office receipts before to a film’s debut is investigated ( Ahmad et al., 2020b ). This method concentrates on early-stage prediction by examining public involvement and sentiment, in contrast to earlier approaches that depend on Twitter or IMDb evaluations after publication. The model presents novel indicators such the like-to-dislike ratio, the positive-to-negative sentiment ratio, and purchase intention. According to experimental data, the suggested approach achieves a relative absolute error of 29.65%, outperforming three baseline models. This strategy demonstrates how pre-release social media interactions may be used to anticipate movie income earlier and with more accuracy.

A WiFi-based sensing method presented to better estimate revenue in retail environments and comprehend consumer behavior ( Golderzahi & Pao, 2024 ). The system finds groups of consumers with similar habits by using WiFi access points placed in cafeterias to track client visiting patterns, including Service Set Identifier (SSID) data. Predictive modeling utilizing machine learning methods such as Random Forest and Support Vector Regression is based on these groupings. Revenue from coffee shops, the quantity of products sold, and the number of consumer devices are the three main outcomes that the models seek to forecast. Predictive accuracy is greatly increased by adding weather and customer group data, with a Mean Absolute Percentage Error (MAPE) improvement of 6% to 10%. This study shows how behavioral data and environmental elements may be effectively integrated to improve retail forecasting accuracy.

Using information from the Ethiopian Commodity Exchange (ECX), it was predicted the future prices of two important Ethiopian export commodities: coffee and sesame ( Fofanah, 2021 ). By using and contrasting three algorithms—Linear Regression (LR), Extreme Gradient Boosting (XGB), and Long Short-Term Memory (LSTM)—it fills a research gap. The study assesses the accuracy of each model using datasets of 7205 for sesame and 1540 for coffee. Ethiopia Coffee Prices Predictor (ECPP), an easy-to-use mobile app, was also created to make price forecasts available, demonstrating the potential of mobile-based forecasting tools in regional commodity markets.

The difficulties in predicting income in the food sector, particularly for restaurants in urban locations, are focused ( Sanjana Rao et al., 2021 ). Three restaurant types—inline, food court, and mobile—are the subject of the study. It suggests a method for forecasting restaurant income that takes into account a number of factors, ranking the input features according to how they affect the desired attribute. To enhance the quality of the dataset, the study uses pre-processing methods such Principal Component Analysis (PCA), feature selection, and label encoding. Following dataset training, the models are assessed, and it is shown that Random Forest (RF) predicts revenue more accurately than Linear Regression. Additionally, the study demonstrates that pre-processing greatly improves model accuracy.

However, there is a need for straightforward, understandable, and useful models made for small businesses like coffee shops because the majority of current research focuses on complicated models or large-scale firms. The purpose of this study is to close the gaps.

Figure 1 illustrates the main architecture of the proposed revenue prediction system. It consists of seven modules. Module 1 describes the dataset, and the second module shows the statistical summary of the dataset. Feature analysis has been explored in the third module. Module 4 demonstrates the relationship of each feature with the target variable. After that 5th and 6th modules calculated the correlation matrix and assisted in selecting the top input features. Finally, in module 7, I have applied the linear regression-based model.

The objective of this study was to develop a prediction model that could anticipate desired outcomes like demand or revenue using structured data and machine learning techniques, especially linear regression. It gave interpretability, practical application, and model correctness top priority throughout the process. Box plots were utilized to examine feature distributions and find outliers, skewed variables, and underlying trends throughout the data exploration stage. Important details on the behavior of particular traits and their possible impact on income were uncovered by this visual analysis. By ranking the most important predictors using feature relevance analysis, we were able to simplify the model and enhance performance. It is clear from Figure 10 that the most important indicators for predicting revenue are the number of customers per day, average order value, and daily marketing spend. Secondary elements like location foot traffic, the number of employees and operating hours come after this.

With a high R² value of 0.895, the chosen linear regression model explained over 89.5% of the variation in daily income. A strong linear association between the chosen characteristics and the target variable is shown by this high explanatory power. The accuracy of the model’s predictions is further supported by the Mean Absolute Error (MAE) of 244.12. From a commercial perspective, these results provide small coffee shop owners with useful, data-driven insights that they can implement. Marketing efforts should prioritize customer acquisition strategies, such as local advertising, social media campaigns, or partnerships with foot-traffic-heavy locations, over initiatives that aim to increase average order value alone, for example, since the number of daily customers had the highest impact on revenue.

Additionally, even while passive location measures show a smaller association, investing in foot traffic visibility enhancements (such as outside signs or marketing) may still pay off. These tactics provide business stakeholders with precise, doable advice and immediately mirror the model’s conclusions.

In a broader sense, this approach facilitates revenue forecasting, demand analysis, and strategic resource allocation. It encourages data-driven decision-making and reduces risks, which is useful not just in the retail industry but also in the restaurant business and even in commodity-based industries like agriculture. Stakeholders may now more accurately forecast changes in income and make proactive adjustments to operations or budgetary allocations, for instance.

Overall, the study shows that it is both possible and efficient to model and forecast important business indicators using supervised learning approaches, particularly linear regression. The robustness and predictive power of the model might be further enhanced for further research by including time-series components, experimenting with ensemble learning techniques, advanced machine learning, or utilizing deep learning models.

This study uses correlation-based feature selection and linear regression to explore how well a data-driven method can predict coffee shop income. This analysis identified the key predictors of daily revenue by examining operational factors such as average order value, marketing expenses, number of employees, and daily customer volume. A substantial amount of the revenue variance can be explained by the linear regression model that was created using these highly correlated characteristics, as seen by its high R² score of 0.89. Further confirming the correctness and dependability of the model were evaluation measures such as MAE. A substantial amount of the income variance can be explained by the linear regression model that was created using these highly correlated characteristics, as seen by its high R² score of 0.89. Assessment indicators such as RMSE and MAE provided additional confirmation of the model’s accuracy and dependability. The results emphasize how crucial order value optimization, customer flow management, and focused marketing campaigns are to increasing revenue. By providing business leaders with practical information, this method helps them make better strategic decisions.

The model was trained and evaluated on a single public dataset, which is a limitation of the study, even if it performs well on the dataset that is currently accessible. As a result, it might not be as applicable in other retail settings. To guarantee wider application, future studies should test the model utilizing private data from various retail establishments and coffee shops. The model’s generalizability is still constrained because the dataset does not reflect multi-location data from coffee shops. Future research should test this model using real or private data from various business kinds and geographical areas.

Future research can improve prediction power by incorporating seasonal trends, consumer segmentation, and external contextual factors like weather, holidays, and seasonal patterns are not taken into consideration in the present dataset. Such characteristics should be taken into account in future research as they have the potential to greatly increase model accuracy, even if the model works well with the information provided. Overall, this work offers a solid basis for using straightforward but powerful machine learning models to anticipate retail sales.