Leveraging ARIMA and Prophet Models for Effective Sales Forecasting and Inventory Management

In today’s fast-paced and data-driven market environments, accurate sales forecasting and efficient inventory optimization are critical for minimizing losses and maximizing profitability. This study explores modern data-driven approaches to enhance customer demand forecasting by intelligently selecting and incorporating relevant market indicators, moving beyond traditional models that rely solely on historical sales data. Utilizing Forward Feature Selection, we automate the identification of impactful variables, improving forecasting accuracy without requiring domain-specific manual input. Experimental evaluation on diverse retail sales datasets demonstrates that the integration of selected features significantly improves the performance of both ARIMA and Neural Prophet models, with up to 23% reduction in forecasting error (MAE and RMSE) compared to baseline models. Furthermore, we extend the orthogonal statistical learning framework to develop causal time series models that better generalize to unseen scenarios and policy interventions. To facilitate practical deployment, a user-friendly, web-based application is introduced, integrating these advanced forecasting tools. The platform enables businesses to make smarter inventory decisions, better respond to market changes, and proactively manage demand fluctuations, ultimately enhancing operational efficiency and customer satisfaction.