Bayesian networks applied to sales forecasting in a large Brazilian fast-food chain

The fast-food segment has become a very competitive market with large companies. Artificial intelligence methods can offer numerous benefits for this market, such as allowing the development of computational models for decision-making. This work aims to develop probabilistic models based on Bayesian Networks to make sales predictions and analyze the causality between variables that influence the sales process of certain groups of products in this segment. We consider the adoption of the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology in the stages of data science, as well as the application of structure learning algorithms in Bayesian Networks from data, and parameter learning. Finally, we performed cross-validation and evaluation based on the Mean Absolute Percentage Error (MAPE) metric. The techniques proposed achieved promising results in terms of forecasting and causality analysis.