ENHANCING SALES PREDICTION FOR SMALL AND MEDIUM-SIZED ENTERPRISES (SMES) USING MACHINE LEARNING

Abstract

This study applies machine learning techniques to develop sales forecasting models for SMEs in the catering sector. Unlike traditional accounting research that relies solely on historical financial variables, our approach integrates both historical financial and non-financial predictors. We discover that including non-financial predictors substantially improves the accuracy of firm level sales forecasts, particularly when employing an advanced machine learning technique, namely Gradient Boosting Regression Tree (GBRT). Additionally, incorporating customer-level sales data enhances the prediction accuracy of firm-level sales, especially when using a multi-target multi-level modeling approach. Our findings underscore the importance of both non-financial data and detailed customer-level sales data in enhancing firm-level sales forecasting. However, realizing the full predictive potential of these data sources necessitates not only the use of sophisticated machine learning methods but also an innovative integrated modeling approach.