The integration of predictive analytics and machine learning for demand forecasting in e-commerce: A theoretical exploration

Abstract

Effective supply chain management hinges on accurate demand forecasting. Yet, traditional methods often struggle with the noise and distortions inherent in communication patterns between supply chain participants. This paper explores the potential of machine learning (ML) to overcome these limitations in the context of e-commerce. We compare the performance of various ML-based forecasting techniques with established methods using data from a chocolate manufacturer, a toner cartridge manufacturer, and the Statistics Canada manufacturing survey. While the overall average accuracy of ML techniques doesn't outperform traditional approaches, a specifically trained support vector machine (SVM) incorporating multiple demand series emerges as the most effective forecasting tool. These findings suggest that, while further research is warranted, strategically leveraging ML holds promise for enhancing e-commerce demand forecasting by learning from complex, noisy data patterns.