Use of machine learning to optimize distribution for Small and Medium Enterprises

Abstract

Small and medium-sized enterprises (SMEs) play a crucial role in Kenya’s economic development, yet they face persistent challenges in managing distribution and inventory efficiently. This study investigates the application of machine learning (ML) techniques to optimize distribution strategies and forecast sales quantity for SMEs, with a focus on the beauty and cosmetics sector. Using the CRISP-DM methodology, the research employs ensemble models—Decision Tree, Random Forest, and Gradient Boosting—to analyze historical sales and distribution data. The Random Forest model achieved the highest predictive accuracy, outperforming other models based on RMSE, MAPE, and R² metrics. Key features such as outlet type, and delivery route cluster emerged as significant predictors of sales quantity. The findings underscore the value of ensemble learning in capturing non-linear relationships and enhancing inventory planning and delivery scheduling for SMEs. This study contributes to theoretical frameworks such as Control Theory and Customer Value Theory by demonstrating how ML supports dynamic feedback and value-based customer segmentation. It also provides actionable recommendations for SMEs, including the digitization of inventory systems, adoption of interpretable ML models, and investment in real-time analytics. Limitations such as data sparsity and lack of external variables are discussed, and future research is encouraged to explore more advanced models, diverse datasets, and real-world deployment in SME environments. Keywords: Distribution Effectiveness, Customer Satisfaction, Sales and Demand Forecasting