Evaluating the Quality of Agglomerative Hierarchical Clustering on Crime Data in Indonesia

Rizkya, Dini Dara
Retno, Sujacka
Yunizar, Zara

International Journal of Engineering, Science and Information Technology
Vol 5, No 2 (2025)

This study evallualtes the quallity of ALgglomeraltive Hieralrchicall Clustering with single linkalge, complete linkalge, alveralge linkalge, alnd walrd linkalge on the daltalset of the number of criminall calses in Indonesial (20ll0ll0ll-20ll23). The analysis compares clustering performance on the original and normalized datasets using the Davies-Bouldin Index (DBI), Silhouette Score (SS), Normalized Mutual Information (NMI), Adjusted Rand Index (ARI), and Callinski-Harabasz Index (CH). The results showed that Ward Linkage provided the best clustering results, with the highest CH increasing from 65.826 to 66.873, clear cluster separation, and a stable structure (NMI = 0.5855, ARI = 0.6298). Single Linkage experienced a chaining effect, although it showed improvement in DBI from 0l.1793 to 0l.1765 and SS from 0l.6271 to 0l.640l0l, with NMI and ARI stable at 0l.4537 and 0l.5865, but CH decreased from 21.731 to 21.0l72. Complete Linkage was too aggressive in separating the data, shown by an increase in DBI from 0.5327 to 0.7116 and a decrease in SS from 0.6336 to 0.5830, although CH increased from 64.244 to 66.873. Average Linkage showed stable results, with NMI = 0l.6481 and ARI = 0l.7993 remaining, but a slight decrease in DBI from 0l.3874 to 0l.40l91, SS from 0l.6839 to 0l.6825, and CH from 42.358 to 40l.251. Data normalization generally helps to improve clustering quality by reducing the influence of feature scale differences. Several metrics showed improved cluster separation on normalized data, although the impact varied depending on the linkage method. Overall, Ward Linkage with normalization is recommended as the best method to produce more accurate clustering in Indonesia's crime data analysis.