OPTIMASI JUMLAH CLUSTER PADA K-MEANS CLUSTERING MENGGUNAKAN PARTICLE SWARM OPTIMIZATION UNTUK PENGELOMPOKAN UKT MAHASISWA

Ira Fazira (Universitas Malikussaleh)
Zahratul Fitri (Universitas Malikussaleh)
Risawandi (Universitas Malikussaleh)

RABIT: Jurnal Teknologi dan Sistem Informasi Univrab
Vol 10 No 2 (2025): Juli

The determination of the Single Tuition Fee (UKT) group in higher education faces challenges in terms of distribution fairness due to the inappropriate grouping of students' socio-economic conditions. The K-Means algorithm, while effective in handling large-scale data at good computational speeds, has a drawback in determining the optimal number of clusters automatically. This study aims to implement the integration of Particle Swarm Optimization (PSO) with K-Means Clustering in the grouping of student UKT data and evaluate the improvement of the quality  of clustering produced compared to conventional methods. The study uses a dataset of 437 new students of the Faculty of Engineering in 2024 from Malikussaleh University with 8 attributes that describe family socioeconomic conditions. The research stages include pre-processing of data, determination of the optimal number of clusters using PSO, implementation of K-Means clustering with optimal K, model evaluation using Silhouette Coefficient and Davies-Bouldin Index, and model comparison using the elbow method. The results of the study showed that PSO succeeded in determining the optimal number of clusters as many as 3 clusters. The implementation of K-Means with K=3 resulted in the distribution of clusters: cluster 0 (40 students/9.2%), cluster 1 (93 students/21.3%), and cluster 2 (304 students/69.6%). Clustering quality evaluation  resulted in  a Silhouette Coefficient of 0.278062 and  a Davies-Bouldin Index of 1.430505 indicating adequate cluster formation with fairly good internal cohesion and reasonable separation between clusters. Comparison with  the conventional K-Means method  using the Elbow Method shows the advantage of PSO-K-Means with  a higher Silhouette Coefficient (0.278062 vs 0.250300) and  a competitive Davies-Bouldin Index (1.430505 vs 1.315400). This research proves that the combination of PSO and K-Means can provide a more optimal solution in the grouping of student UKT to support a fairer determination of tuition fees based on family economic ability.