Machine Learning-Based Customer Demand and Online Sales Prediction

Abstract

The development of Internet technology has changed how customers engage with e-commerce websites, where transactional and customer-related data may offer insightful information on consumer behavior. Companies must handle massive amounts of multivariate data to enhance demand analysis and provide efficient services in response to growing consumer expectations and fierce online competition. The e-commerce sales forecasting problem is addressed in this study, based on actual transactional, temporal, logistical, pricing, payment, and review data from the Olist Brazilian E-Commerce Dataset. The proposed methodology uses data preprocessing techniques such as missing value treatment, date-time transformation, duplicate removal, label encoding, Z-score standardization, and SMOTE-based data balancing. A complex sequential pattern learning model, called Long Short-Term Memory (LSTM) is built to learn complex sequential patterns and predict customer demands and online sales. The experimental results have shown that the proposed LSTM has been able to attain 99.72% accuracy, 99.64% precision, 99.16% recall, and 99.26% F1-score. The results show that the proposed methodology is useful in analyzing e-commerce demand and forecasting e-commerce sales. The model can help businesses manage their inventory, sales, resources, and decision-making processes.