Enhancing Sales Prediction Accuracy Through Incorporation of Expert Knowledge as Constraints Into the Model

Data-driven prediction models are increasingly gaining importance in business decision-making processes. These models enable the identification of future trends and potential opportunities by making predictions based on historical data. However, to enhance the accuracy and usefulness of these predictions, expert opinions and business processes need to be integrated into the model. This study addresses the development of a prediction support system and the integration of user guidance capabilities for a cold chain food supplier company. We have incorporated external data sources, including weather conditions and public holidays, alongside the company's invoice data spanning from 2017 to 2023. A key focus of our research is the implementation of constraints based on domain knowledge and expert opinions to improve model accuracy. We developed a hybrid prediction model that combines data-driven learning with expert-guided constraints. Our results demonstrate a significant improvement in prediction accuracy.