Driven by the rapid advancement of the internet and e–commerce, digital marketplaces have emerged as pivotal pillars of modern commerce. In an intensely competitive e–commerce landscape, precise demand forecasting and agile operational strategies are critical to bolstering enterprises’ market competitiveness. This study leverages Python to develop a hybrid forecasting system that integrates the Prophet time–series model and the random forest algorithm, addressing intricate forecasting challenges inherent in e–commerce datasets. The research is grounded in a comprehensive data analytics workflow, utilizing a public dataset comprising 232,620 Taobao transaction records from 2023. The model incorporates a dynamic weighting mechanism: the random forest algorithm is deployed for long–term forecasting to capture the nonlinear dynamics of user behavior. For user behavior analysis, heatmap visualization identifies variables strongly correlated with sales volume. These external variables are integrated into the random forest model for residual prediction. A comparative analysis of residuals generated by the hybrid model versus the standalone Prophet model is conducted. This study delivers both theoretical and technical contributions. Theoretically, it unifies time series analysis and machine learning methodologies to construct a quantitative model linking behavioral characteristics to sales responses. Technically, it enhances model training efficiency through sliding–window aggregation and other optimizations. Limitations include a dataset focused on 3C product categories, slow responsiveness to black swan events, and suboptimal real–time performance. Future work will advance real–time forecasting systems through multimodal data fusion and extend research to emerging scenarios such as cross–border e–commerce, fostering the intelligent and scalable evolution of e–commerce forecasting technologies.