Customer behavior Analysis for E-Commerce and Retail using Improved K-Means Clustering with Calinski Harabasz Index

In recent days, the customer behavior understanding is crucial for business to tailor the marketing strategies in the rapid growth of e-commerce landscapes to improve customer satisfaction. However, the traditional clustering algorithms struggled with cold start, scalability and sparsity issues which led to inaccurate results. Therefore, this research proposes an Improved K-Means Clustering with Calinski Harabasz Index (IMC-CHI) model for behavior analysis of customer in e-commerce. The K-means clustering algorithm assigns weights to the significant features to improve the clustering of new or sparse data. In addition, the CHI determines the optimal number of clusters by ensuring efficient and scalable customer behavior segmentation. Initially, the data is collected from Brazilian E-Commerce Public Dataset by Olist (BECPDO) and further preprocessed to handle missing values with K-Nearest Neighbor (KNN) imputation. Then, the recency, frequency, and monetary indicators are extracted and normalized with z-score normalization to avoid disparity. Finally, the proposed IMC-CHI is introduced to cluster the customer behaviors of e-commerce landscapes. From the results, the proposed IMC-CHI outperformed Improved K-Medoids (IKM) algorithm in terms of CH as 8100 and Sum of Squared Errors (SSE) as 34000 per 12 clusters respectively.