The retail industry increasingly relies on data-driven decision-making to maintain competitiveness in a rapidly evolving market. However, integrating and analyzing large, diverse datasets remains challenging due to issues like data fragmentation, high dimensionality, and varied data types. This study addresses these obstacles by examining the application of clustering algorithms, specifically K-means and its enhanced variant K-means++, within a distributed computing framework. This research aims to develop scalable and efficient methods for transforming disparate retail data into actionable insights. This research proposes an integrated approach combining K-means++ with the Apache Spark platform to enhance clustering performance on retail datasets. The methodology includes optimizing centroid initialization using a probability-based selection process and implementing distributed computation for handling large volumes of data. The study further explores practical applications, including customer segmentation, anomaly detection, and geo-analytical evaluation in retail contexts. Results indicate that the K-means++ algorithm, especially when accelerated through Spark, significantly improves clustering stability and efficiency compared to standard K-means. It effectively supports the identification of meaningful customer groups and operational patterns within complex retail data. However, limitations persist regarding its sensitivity to outliers, inability to handle categorical data natively, and limited adaptability to non-spherical cluster structures. Future work should incorporate hybrid models and explainable AI techniques to enhance interpretability and generalizability across diverse retail environments.
Miao Yan (Wed,) studied this question.