Grey Wolf Algorithm For Optimizing K-Modes in Determining Initial Centroid

Clustering categorical data presents unique challenges due to the difficulty of measuring distances between categorical variables. The K-Modes algorithm is widely used for categorical clustering; however, it struggles with determining initial centroids, which can impact clustering performance. This study proposes using the Grey Wolf Optimization (GWO) algorithm to improve K-Modes by optimizing initial centroid selection. Utilizing data from the 2021 Village Potential Survey conducted by the Indonesian government, a comparative study was conducted between the conventional K-Modes and K-Modes integrated with Gray Wolf Optimization (GWO) to evaluate their clustering effectiveness. The findings indicated that K-Modes+GWO consistently improved silhouette scores across various cluster sizes, with the most significant enhancement observed in the two-cluster configuration, where the silhouette score rose from 0.2264 to 0.3049. Furthermore, the Davies-Bouldin index decreased from 1.8932 to 1.3783 in the same configuration, suggesting that clusters became more compact and distinct. Visual assessment also demonstrated better demarcation and clarity of cluster boundaries, especially notable in the two-cluster case. These results underscore the capability of optimization algorithms like GWO to enhance clustering quality for nonnumerical data, offering a valuable approach to improving clustering methodologies for categorical data.