Forecasting Sales at Fuel Stations Using a Multilayer Perceptron

Forecasting fuel sales at gas stations is critical for effectively managing and optimizing business processes. Despite the importance of sales forecasting, many gas stations still rely on traditional methods such as analyzing historical short-term sales data or intuitive managers’ judgments. However, there is a need to use advanced forecasting methods, such as deep learning, which can take into account complex patterns and relationships between various factors influencing demand. The aim of this paper is to develop a deep learning method for sales forecasting at fuel stations based on the multilayer perceptron (MLP) and optimization of the developed MLP structure. The results of the research show that the performance of multilayer perceptron is better than the performance of linear regression. However, the effectiveness of the methods for forecasting sales at gas stations using deep learning proposed in this work depends on the amount of data entered as well as on the limitations provided in the MLP model.