Applying Machine Learning Algorithms to Solve the Data Imbalance Problem in Predicting the Effectiveness of E-Commerce Advertising for Small and Medium-Sized Enterprises in Vietnam

Abstract

In Vietnam, SMEs (small and medium-sized enterprises) or MSMEs (micro, small and medium-sized enterprises) usually struggle to predict the effectiveness of advertising on e-commerce platforms using machine learning. When the input data on successful campaigns only accounts for a small fraction, the dataset is imbalanced, that easily leading to biased results based on the majority class. This study uses a real-world dataset from a cosmetics store on Shopee. We evaluate the dataset across eight machine learning (ML) models combined with advanced oversampling techniques such as Borderline-SMOTE, ADASYN, and SMOTEENN. Experimental results indicate that combining Borderline-SMOTE with neural networks provides balanced performance, achieving a recall accuracy of 0.4545 and an F1 score of 0.2143. Research confirms that addressing data imbalances is crucial for SMEs when applying machine learning-based ad prediction. When SMEs adopt this method, they can optimize marketing operations and improve return on investment.