Evaluation of K-Means Clustering Using Silhouette Score Method on Customer Segmentation

Customer segmentation is a critical process in businesses to understand and meet the diverse needs of customer. This study focused on the challenges of managing large and complex volumes of customer data and identifying the right segments to personalize marketing strategieshow about if I . K-Means Clustering has been widely utilized for its ability to group multidimensional data, but this method often generated broad clusters that lack detailed insights. Therefore, cluster evaluation with the Silhouette Score method became essential to ensure the optimality and validity of the generated groupings. The purpose of this study was to evaluate the quality of K-Means Clustering using the Silhouette Score method on customer segmentation. This research began with the acquisition of a dataset comprising 2,000 data points characterized with 7 attributes: sex, marital status, age, education, income, occupation, and settlement size. The data then underwent pre-processing by checking missing values and normalizing data. K-Means Clustering was then applied to group data into several clusters based on their proximity to the cluster center (centroid). The results of the clusters were assessed using the Silhouette Score method to determine the most optimal number of clusters. The results of this study consisted of manual calculations using Microsoft Excel on 27 data points to facilitate understanding of the logic, steps, methods and practical foundations before implementation on the complete dataset. Furthermore, the results of the Python calculation in 2000 data points showed that the optimal number of clusters (close to the value of 1) between k = 2 to k = 7 was the k = 4 cluster with a Silhouette Score value of 0.43, categorized as a weak structure. Although this value indicated a weak cluster structure, it was the highest value in the test, indicating that the division of data into four clusters (k = 4) was better than the number of other clusters. However, the quality of this cluster indicates the need for futher improvement. Future work should review the used attributes, data normalization methods, or consider other clustering algorithms to achieve a more robust structure and more meaningful interpretation.