Striking a balance: navigating the trade-offs between predictive accuracy and interpretability in machine learning models

Sales forecasting is very important in retail management. It helps with decisions about inventory, staffing, and planning promotions. In this study, we looked at how to balance the accuracy of predictions with how easy it is to understand the machine learning models used in sales forecasting. We used public data from Rossmann stores to study various factors like promotions, holidays, and store features that affect daily sales. We compared a complex, highly accurate model (XGBoost) with simpler, easier-to-understand linear regression models. To find a middle ground, we created a hybrid model called LR_XGBoost. This model changes a linear regression model to match the predictions of XGBoost. The hybrid model keeps the strong predictive power of complex models but makes the results easier to understand, which is important for making decisions in retail. Our study shows that our hybrid model offers a good balance, providing reliable sales forecasts with more transparency than standard linear regression. This makes it a valuable tool for retail managers who need accurate forecasts and a clear understanding of what influences sales. The model's consistent performance across datasets also suggests it can be used in various retail settings to improve efficiency and help with strategic decisions.