Perbandingan Clustering Saham dengan Metode BHC, PAM, dan DBSCAN

Abstract

Penelitian ini bertujuan untuk membandingkan kinerja tiga metode clustering, yaitu Bayesian Hierarchical Clustering (BHC), Partitioning Around Medoids (PAM), dan Density-Based Spatial Clustering of Applications with Noise (DBSCAN), dalam mengelompokkan saham berdasarkan indikator keuangan fundamental. Data yang digunakan berupa 808 saham yang diperdagangkan di Bursa Efek Indonesia (BEI) pada tahun 2024, dengan variabel meliputi beta saham, return saham, Return on Asset (ROA), Return on Equity (ROE), dan Earning per Share (EPS). Metode BHC menghasilkan empat cluster, metode PAM menghasilkan dua cluster, sementara DBSCAN membentuk dua cluster utama. Hasil evaluasi menggunakan silhouette score menunjukkan bahwa metode DBSCAN memiliki performa clustering yang lebih baik dengan nilai rata-rata sebesar 0.66, metode PAM menghasilkan nilai 0.51, sedangkan metode BHC hanya menghasilkan nilai -0.02. Analisis karakteristik cluster menunjukkan bahwa DBSCAN mampu mengelompokkan saham secara lebih jelas dan efisien berdasarkan pola kinerja keuangan. Oleh karena itu, DBSCAN lebih direkomendasikan sebagai metode clustering saham dalam konteks data yang bersifat kontinu dan memiliki outlier.

Kata kunci: BHC, clustering, DBSCAN, PAM, silhouette score.

This study aims to compare the performance of two clustering methods, namely Bayesian Hierarchical Clustering (BHC), Partitioning Around Medoids (PAM), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), in classifying stocks based on fundamental financial indicators. The dataset consists of 808 stocks traded on the Indonesia Stock Exchange (IDX) in 2024, with variables including stock beta, stock return, Return on Assets (ROA), Return on Equity (ROE), and Earning per Share (EPS). The BHC method produced four clusters, PAM formed two main clusters, while DBSCAN formed two main clusters. Evaluation using the silhouette score shows that DBSCAN has better clustering performance with an average score of 0.66, PAM achieved 0.51, while BHC only achieved -0.02. The cluster characteristic analysis demonstrates that DBSCAN is more capable of grouping stocks clearly and efficiently based on financial performance patterns. Therefore, DBSCAN is more recommended as a stock clustering method, especially for continuous data with potential outliers

Keywords: BHC, clustering, DBSCAN, PAM, silhouette score.