<![CDATA[Evaluating the effectiveness of government training programs requires more than a binary employed/unemployed indicator, since participants who find work still differ substantially in salary level, waiting time to employment, job position, gender, and employment sector. Prior studies applying clustering to workforce or socio-economic data have generally compared only two or three algorithms at a time, on datasets from other domains such as banking or health, leaving it unclear which method best segments multidimensional post-training outcome data with mixed numerical and categorical attributes and no natural hard boundaries between groups. This study addresses that gap by comparing four clustering algorithms, K-Means, Hierarchical Clustering, Gaussian Mixture Model, and Fuzzy C-Means, to segment post-training participant data from the Surabaya City Government, with the aim of providing local policymakers with an evidence-based grouping of participants that can inform which sectors and job levels most need follow-up support. The dataset consists of 509 observations with attributes for job position, monthly salary, gender, employment sector, and waiting time. Prior to clustering, the data were preprocessed through cleaning of inconsistent categorical entries, median imputation of missing numerical values, One-Hot Encoding of categorical attributes, and Min-Max normalization of numerical attributes to a common 0-1 scale. Clustering performance was evaluated using Silhouette Score and Davies-Bouldin Index (DBI) across cluster counts k = 2-5. Fuzzy C-Means with five clusters achieved the best overall performance, with a Silhouette Score of 0.649 and a DBI of 0.640; its closest competitor, Ward-linkage Hierarchical Clustering, achieved a comparable Silhouette Score of 0.647 but a markedly higher DBI of 0.872, indicating that FCM produced more compact, well-separated clusters overall even though the two methods separated participants almost equally well. The resulting five clusters show clear policy-relevant differences: the largest cluster, dominated by the industrial sector, has the highest average salary but also the longest waiting time to employment, suggesting that industrial-sector training would benefit from faster competency certification and job-matching support, while smaller, female-dominated clusters in the social sector show shorter waiting times but lower salaries, pointing to a need for upskilling pathways into higher-paying roles. These findings illustrate how clustering can move post-training evaluation beyond simple placement rates toward data-driven, sector-specific recommendations for workforce training policy.]]>
