Sales Performance Classification of Promotional Products Using Data Mining

Abstract

The objective of this research is to establish a classification model to determine high sales performance promotional products using past sales records. The issue at stake is that it is very hard for business actors to forecast the appearance of high sales promotional products, taking into account different factors, such as product type, price per unit, quantity requested, and sales period. This study, based on a quantitative and experiential manner, makes use of the C4. 5 decision tree algorithms on real transaction data of HERA Promotion during 2024. The data falls into one of two types: "best-selling" and "not-selling" products. The proposed classification model achieved good generalization performance with the test accuracy of 99.48% and 5-fold cross-validation accuracy of 96.77%. Price Unit, Product Name, and Month were the most essential features in classifying products, showing that economic value and seasonal demand are major factors determining whether a product is sold. But when applied to the external data for the first few months of 2025, accuracy fell to 78%, which is a way for them to show that shifts in consumer behavior can drive changes in performance. From a theoretical point of view, this study fills the gap by incorporating the time effects as a dynamic variable into product classification models, which haven't been mentioned much in previous research. For practice, the results also encourage incorporating data-driven classification models within decision support systems to help with stock planning and promotional strategies. More work is warranted to use ensembling techniques and real-time data streams on the generalization ability and adaptability of the models.