Sales prediction hybrid models for retails using promotional pricing strategy as a key demand driver

Abstract

The implementation of promotional pricing strategies constitutes a key component within the realm of retail revenue management. Nonetheless, the accurate prediction of sales in the presence of price discounts proves challenging due to the influence of various factors that contribute to demand uncertainty and high fluctuations. This study aims to find the most suitable prediction models for retail product unit sales while comprehensively accounting for the complex impacts of contributing factors. The dataset, sourced from a case study of a retail company, spans the temporal interval from January 2020 to December 2022. The predictive models, encompassing linear regression, random forest, XGBoost, artificial neural networks, and hybrid machine-learning models, are systematically developed. Then, the identification of the most suitable model is facilitated through the computation and comparative analysis of the Mean Absolute Percentage Error, with due consideration given to the weighting by the respective product’s revenue, thereby offering a comprehensive assessment of overall performance. Additionally, different types of feature selection are experimented. Factors used in machine learning models are either using all the independent variables or using significant factors from the stepwise method, and either considering or not considering exogenous factors of other products in the same cluster grouped by category, subcategory, or K-means method. The result shows that the series hybrid model of random forest and XGBoost outperformed others. Considering factors affecting sales, it is found that the promotion period factor was the most important, followed by discount percentage and price factors. This research provides analytics framework for sales prediction for retails using promotional pricing as a key demand driver.