Price Trend Prediction and Product Recommendation Using Machine Learning Approach

Abstract

Purpose: This study aims to develop an accurate and relevant machine learning-based model for price trend prediction and product recommendation using retail sales data, in order to support operational efficiency and marketing strategies. Design/methodology/approach: The study utilizes sales data from two supermarkets spanning the years 2021– 2024. Five machine learning algorithms are compared, namely Random Forest, XGBoost, Gradient Boosting, Decision Tree, and Linear Regression, with evaluation metrics including R², MAPE, and RMSE. The best- performing model is then optimized using feature importance and feature selection methods. For the recommendation system, a Content-Based Filtering approach is employed based on similarity in product names and categories. Findings/resultWith an R2 of 80%, MAPE of 25%, and RMSE of 1.07 at the start, the Random Forest model performed the best. Following feature optimization, it greatly improved, reaching R2 of 92%, MAPE of 13%, and
RMSE of 0.84. The Content-Based Filtering recommendation system successfully provided relevant suggestions, with the highest cosine similarity reaching 0.4268. Originality/value/state of the art: This study is grounded in and extends the findings of previous research.