<![CDATA[Daily sales forecasting is a critical component of business planning that must adapt to the dynamics of market demand. While traditional approaches such as Single Moving Average and Trend Moment have been used in previous studies, their predictive accuracy on daily sales often remains suboptimal, with reported MAPE values up to 39.2%. Prophet, a time series model developed by Meta, offers enhanced flexibility in capturing non-linear trends, seasonality, and incorporating external regressors. This study proposes a hybrid forecasting model by combining Prophet with engineered features and external regressors, including calendar effects and recent sales statistics. The dataset consists of daily sales records that have undergone data cleaning, logarithmic transformation, and smoothing. Prophet is configured with additional monthly seasonality, national holiday indicators, and optimized parameters through grid search. Evaluation results demonstrate a substantial improvement, with the final model achieving an R² score of 0.9787 and a MAPE of 3.79%, outperforming conventional methods and aligning with the best results from recent Prophet-based studies. These findings confirm that the integration of external variables within Prophet significantly improves prediction accuracy, making it suitable for time series forecasting in various business domains with similar data patterns.]]>
Copyrights © 2025
