Accurate demand forecasting is a critical requirement for supply chain optimization, inventory control, and strategic decision-making in modern enterprises. Traditional statistical models, while widely used, often fail to capture the nonlinear dynamics and sudden variations in consumer behavior, leading to inefficiencies in planning and resource allocation. This paper investigates the application of machine learning algorithms to product demand forecasting by integrating historical sales records with factors influencing purchasing decisions. A comparative analysis is conducted between classical time-series methods (ARIMA, SARIMA, Prophet) and advanced machine learning approaches (Random Forest, XGBoost, and Long Short-Term Memory networks). Empirical evidence indicates that machine learning techniques outperform conventional methodologies in terms of stability and predictive precision. Specifically, LSTM architectures exhibit superior reliability when managing complex, long-term data correlations. Utilizing these data-centric forecasting tools enables organizations to minimize expenditures, streamline operations, and facilitate enduring corporate expansion. The study offers practical data-driven recommendations and visualization tools, bridging the gap between advanced predictive modeling and managerial decision-making. Future work should focus on integrating external macroeconomic factors and exploring emerging transformer-based architectures for enhanced scalability.
Yerassyl et al. (Tue,) studied this question.