Machine Learning-Based Decision Support System for Optimizing Business Operations and ResourceAllocation

Abstract

Traditional managerial heuristics like moving averages are widely used in operational decision-making subject to uncertainty, and they are slow to respond to volatility, which results in misallocation of resources, surplus of inventory, and capital wastage. While digital transaction data can help inform evidence-based planning, research in organizations has shown that there is a continuing lack of predictive analytics to produce better, cost-effective decision results. To overcome this challenge, this research introduces and tests a machine learning-enabled decision support system (DSS) which successfully transforms the demand forecasts into an optimized and budget-constrained resource allocation. The methodology is a controlled, repeatable one where a naive baseline rule and a moving average baseline rule are compared to a linear regression, a random forest and a gradient boosting learner, and this comparison holds the allocation rule constant to isolate the effect of the demand signal, while varying the level of resource scarcity. The empirical results show that the random forest-based DSS can significantly improve the accuracy of the forecast, and the mean absolute error and weighted absolute percentage error are lower than the heuristic methods used as a baseline. The system reduces excess stock by 10% to 19% in every budget scenario and achieves reduced stockout rates in balanced and generous situations in resource allocation outcomes. Moreover, the results demonstrate that the value of the ML-based decision support depends significantly on resource scarcity, so that, when budgets are limited, the added value of the ML-based decision support is at its maximum. This study offers helpful benchmarks and strategic direction on how to enhance operational efficiency by treating the DSS as augmentation for the normal replenishment process instead of a complete automation solution.