Forecasting Fuel Retail Sales Volume Using Machine Learning for Sustainable Decision-Making

Purpose: Forecasting fuel retail sales volume is important for sustainable decision-making and process management. Prediction of the petrol products sales volume influences the planning of the effective distribution of fuels. Therefore, it impacts all other supply chains and each country's economy. Method: This research uses machine learning to forecast fuel retail sales volume. The choice of this method and the research subject are based on the complexity of the decision-making process in energy carriers' sales. Machine learning model (LightGBM) and two other state-of-the-art models (ARIMA, Holt-Winters) were compared. Performed calculations are based on real-world data from single fuel company distributors. Results: The three machine learning models are proposed and examined to address this research gap. There is a baseline model proposed along with state-of-the-art models and the results are based on comparative analysis of MSE and RMSE. Finally, a real-world sales data set from a fuel dispenser is analyzed to determine the veracity of the model argument. Conclusion: The best results were obtained by the LightGBM model, utilizing both future covariates and forecasted fuel prices. This model demonstrated the best performance according to all metrics. Performed experiments proved that the best model can support strategic reasoning in organizations interested in forecasting fuel sales. The use of machine learning to forecast retail fuel sales volume for sustainable decision-making is a good example of leveraging modern technology to optimize business processes and make more informed decisions.