This thesis explores a pricing recommendation strategy built for a convenience store chain for the franchisees. The objective is to build a hybrid classification between business logic and machine learning to construct a classification model of the store's product range based on sales and profit. The process starts with business understanding and ends with model evaluation, following the CRISP-DM methodology. Initially, a manual and arbitrary classification was created, where a score based on static thresholds would classify the product type using regular (non-promotional) sales quantity and the franchisee's profit margin. Despite this, this approach has limitations: it is subjective, static, and may not adapt to future market changes. Machine learning overcomesthese limitations by integrating algorithmssuch as Random Forest, KNN and Naive Bayes for validation and to automate classifications. To train and build this classification, 2024 sales data were collected, including margins, prices, and sales across all stores, to study product behavior and classify them strategically into essential, medium and premium. By classifying through algorithms and with the learned models and accurate results, product classifications will allow the pricing strategy to become more automated and to better respond to changes in demand. By associating these classifications with differentiated pricing strategies, the model strengthens the effectiveness of commercial decisions in a dynamic retail context.

Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business Analytics