Revolutionizing Retail Analytics: Advancing Inventory and Customer Insight with AI

Addressing critical issues in retail such as ineffective queue management, inaccurate demand forecasting, and suboptimal marketing strategies, this paper presents a novel solution using advanced machine learning techniques. The proposed Smart Retail Analytics System (SRAS) leverages these technologies to improve operational efficiency and enhance customer engagement. To track customers more effectively, we introduce a hybrid architecture integrating several predictive models. Initially, we fine-tuned the YOLOV8 algorithm with a variety of parameters, achieving remarkable results on key performance metrics using real surveillance footage from retail settings. Subsequently, we incorporated two advanced object-tracking models, BOT-SORT and Byte Track, with YOLOV8's detected labels. This combination is vital for monitoring customer movements within stores, generating precise visitor counts and heat maps, which provide valuable insights into consumer behavior and optimize store operations. For inventory management, we examined and optimized various predictive models against complex retail data patterns. The GRU model, known for interpreting time-series data with long-term dependencies, consistently outperformed others like Linear Regression, showing improvement of 29.3% in mAPE and 2.9% in R2-score.