Enhancing Supply Chain Resilience in Developing Country SMEs through Machine Learning-Based Demand Forecasting: An Empirical Study

Abstract

Small and medium-sized enterprises (SMEs) in developing countries face unprecedented supply chain vulnerabilities due to limited resources, infrastructure constraints, and market volatility. This empirical study investigates how machine learning-based demand forecasting can enhance supply chain resilience in developing country SMEs. Through a mixed-methods approach involving 127 manufacturing SMEs across Pakistan, Bangladesh, and Kenya, we examine the implementation and impact of three ML algorithms: ARIMA, Random Forest, and LSTM neural networks. Our findings reveal that ML-based forecasting reduces demand forecast error by 34-48% compared to traditional methods, decreases inventory costs by 22-31%, and significantly improves order fulfillment rates. However, implementation barriers including data scarcity, technical expertise gaps, and infrastructure limitations persist. We propose a practical implementation framework tailored to resource-constrained environments and demonstrate that even simplified ML approaches yield substantial resilience benefits. This research contributes to supply chain management literature by bridging the gap between advanced analytics capabilities and practical application in developing economy contexts.