<![CDATA[Micro, small, and medium enterprises (MSMEs) in the food and beverage sector typically possess daily transaction records, yet these data are often used only for bookkeeping. As a result, the drivers of sales fluctuations are not clearly identified and operational decisions tend to rely on intuition. This study aims to (1) analyze the key factors influencing daily sales performance of MSME MW and (2) develop an accurate and interpretable sales forecasting model. The main problem addressed is how to transform time series transaction data, which are affected by seasonality, changes in customer behavior, and shifts across digital channels, into informative predictors that can capture non-linear relationships while still providing actionable explanations. The proposed solution integrates time series feature engineering to construct 21 predictive features from POS and digital channel data spanning 19 months, applies a Gradient Boosting Regressor to model complex patterns and improve predictive accuracy, and employs Explainable AI using the SHAP method to quantify both global and local feature contributions to the model output. Preliminary results indicate strong forecasting performance, achieving an R² of 0.987265, an MAE of IDR 33,194.014, and an RMSE of IDR 42,625.808, suggesting a high level of agreement between predicted and actual daily sales. The SHAP analysis identifies “Total Items Sold” and “ATV” as the most dominant drivers of sales increases, while lag-based sales features exhibit non-linear behavior consistent with mean reversion following extreme spikes. Linguistically, these findings imply that revenue growth is driven not only by sales volume, but also by optimizing average transaction value, providing a clear direction for data-driven operational strategies.]]>
