Application of Windowing Techniques to Retail Data for Anomaly Detection
Main Article Content
Abstract
With the growth of retail settings towards data-driven settings, real-time anomaly detection has become an imperative aspect of operational intelligence. Retail data that includes point of sale transactions and clickstream behavior, inventory and sensor logs of IoT devices are more dynamic and high frequency by nature. It is known as windowing and has been used very efficiently by professionals to enable the detection of anomalies in particular cases by a timely and accurate method through segmentation of streaming data into manageable streams. The paper analyzes the uses of different windowing strategies tumbling, sliding, count based, session, and adaptive windows to retail data streams and their effects on the performance of anomaly detection. This paper is based on the synthesis of the main research works, experimental comparisons, and case studies to determine the most successful combinations of windowing techniques and detection algorithms. It also points at research gaps of today and suggests future directions, such as the necessity of adaptive and privacy conscious models. The review would act as a guide to scholars and professionals who are interested in improving anomaly detection systems in retail using smart stream segmentation.