Optimized Decision-Making in Physical Retail Stores Through an Adaptive Hybrid System

Abstract

Physical retail stores have to deliver product recommendations to enhance customer experience, and accurately forecast sales to optimize inventory levels. While traditional recommender systems typically rely on explicit customer preferences, behavior similarities, and popularity metrics, such systems encounter limitations in offline retail contexts due to insufficient explicit customer data and the complexity of customer-store interactions. This paper introduces a novel hybrid solution that combines various recommendation strategies with advanced predictive analytics, including Recurrent Neural Networks and statistical modeling approaches. This approach is designed to improve both the precision of product recommendations and the accuracy of sales predictions. The system was extensively evaluated on a publicly available real-world dataset, using metrics such as Precision@K and NDCG@K to assess its performance. The results demonstrate that the proposed hybrid method successfully balances personalization and accuracy, offering a robust solution for optimized decision-making in physical retail settings.