ADDRESSING CLASS AND DEMOGRAPHIC IMBALANCE IN E-COMMERCE BEHAVIOR PREDICTION: A CASE STUDY USING SMOTE AND RESAMPLING TECHNIQUES
Abstrak
In e-commerce predictive modeling, imbalanced data is a common challenge especially when both class and demographic attributes are unequally distributed. This study explores the combination of Synthetic Minority Oversampling Technique (SMOTE) and demographic resampling to improve the performance of models predicting online purchasing behavior in Malaysia. The first step is to use SMOTE to handle class imbalance in the five-point purchase intention scale classification. The second step is followed by gender imbalance, which is caused by a higher number of female respondents than male, this problem is solved through a combination of down-sampling and up-sampling. Further adjustments were made to balance other demographic factors such as age, employment status, and ethnicity. For this study the dataset, encompassing 1,126 survey responses, was analyzed using WEKA tools with six classifiers: J48, Random Tree, REPTree, JRip, PART, and OneR. Each classifier was evaluated in three stages: original data, after SMOTE, and after both SMOTE and demographic balancing. The results displayed clear improvements in model performance. For example, J48’s accuracy increases from 62.85% (unbalanced) to 98.69% (fully balanced), while Random Tree reached 99.29%. These results highlight the importance of addressing both class and demographic imbalances to develop better and more reliable models. The study improves to the limited research that combines SMOTE with demographic resampling in Malaysian e-commerce and offers practical insights for building better predictive models to support customer segmentation, targeting, and personalization. Future work could explore balancing more attributes and applying the method to ensemble or deep learning models for improved performance and interpretability.
Identifier
- OpenAlex: https://openalex.org/W7161955844
- DOI: https://doi.org/10.24191/mjoc.vo11i1.8077