Інформаційна технологія предиктивної та прескриптивної аналітики для оптимізації виробничого планування

Abstract

The article addresses a relevant scientific and applied problem of improving the efficiency of production planning under demand uncertainty and resource constraints. An integrated approach to decision support is proposed, combining predictive and prescriptive analytics methods within a unified software solution. The architecture of the information technology system was developed to ensure an automated transition from demand forecasting to the formulation of an optimal production plan. A hybrid approach to demand forecasting was implemented, based on the parallel use of statistical ARIMA models and the Random Forest ensemble algorithm. To enhance forecasting accuracy, advanced feature engineering mechanisms were applied, taking into account temporal dependencies as well as exogenous factors, including marketing promotions and calendar events. The selection of the optimal model is performed automatically based on the Mean Absolute Percentage Error (MAPE) metric. A mathematical model for production plan optimization was developed, formulated as a linear programming problem, which considers production costs, inventory holding costs, shortage penalties, and capacity constraints. The integration of forecasted demand into the optimization model allows the implementation of prescriptive analytics concepts and ensures the generation of economically justified managerial decisions. Experimental validation of the proposed approach was conducted using realworld retail network data. The results demonstrated that the use of ensemble machine learning methods ensures a Mean Absolute Percentage Error below 9%. Applying the optimization model provides a reduction of total enterprise costs by 14.5% compared to traditional planning approaches. The practical significance of this work lies in the possibility of using the developed software application as a decision support tool for small and medium-sized enterprises. Future research directions include extending the model to multi-product systems, accounting for stochastic demand, and applying deep learning methods.