COMPARATIVE ANALYSIS OF NEURAL NETWORK AND REGRESSION MODELS FOR CONSUMER DEMAND FORECASTING

Effective consumer demand forecasting plays a crucial role in inventory management, sales planning, and logistics optimization, particularly in the context of economic instability and high variability in consumer behavior. The purpose of this article is to conduct a comparative analysis of the traditional statistical model Prophet and the neural network model LSTM, based on real retail sales data. The methodological basis of the study includes a description of the algorithms, forecasting quality evaluation metrics, and experimental testing using MAE, RMSE, and MAPE indicators. The results show that deep learning models demonstrate higher accuracy under complex demand dynamics, while regression approaches retain advantages in interpretability, deployment speed, and robustness in data-limited environments. The article provides practical recommendations for selecting forecasting models depending on data structure, business objectives, and the company's level of digital maturity. It concludes that hybrid solutions and phased implementation of intelligent forecasting systems are advisable for enhancing demand management in modern business practice.