Electronic Theses and Dissertation

Era digital menyebabkan dunia menghasilkan data dalam jumlah yang semakin besar dan kompleks, sehingga membutuhkan pemrosesan data yang lebih canggih. Evaluasi metode clustering penting dilakukan agar memperoleh metode optimal dalam memproses suatu data. Data penyakit menular disetiap provinsi di Indonesia memiliki pola yang berbeda-beda. Sehingga pengelompokkan provinsi berdasarkan karakteristik penyebaran penyakit menular dilakukan agar strategi penanggulangan dapat disesuaikan dengan kebutuhan setiap provinsi. Penelitian ini bertujuan untuk mengevaluasi metode Partitional Based Clustering dan Density Based Clustering dalam pengelompokan provinsi-provinsi di Indonesia berdasarkan penyakit menular langsung untuk memfasilitasi intervensi kesehatan setiap provinsi. Penelitian ini menggunakan data penyakit menular langsung dari profil kesehatan Indonesia tahun 2022. Analisis komparatif dilakukan dengan menggunakan metode Partitional Based Clustering (K-Means dan CLARA) dan metode Density Based Clustering (DBSCAN dan OPTICS). Pemilihan cluster optimal pada partitional based clustering menggunakan metode Gap Statistics, Elbow, dan Silhouette. Pemilihan MinPts dan epsilon terbaik menggunakan metode KNN. Pendekatan optimal ditentukan berdasarkan metode evaluasi clustering yaitu indeks Calinski Harabasz, Sum of Square Between-cluster, dan indeks C. Pendekatan Density Based Clustering menghasilkan nilai yang sama untuk kedua metodenya, yaitu nilai indeks Calinski Harabasz terbesar senilai 3.624,245, nilai indeks Sum of Square Between-cluster terbesar senilai 72,131 dan indeks C terkecil senilai 0,002 yang mengindikasikan nilai optimal. Pengelompokan optimal mengelompokkan provinsi ke dalam 3 cluster, Cluster 1 terdiri dari 23 provinsi, Cluster 2 terdiri dari 2 provinsi, dan Cluster 3 (noise) terdiri dari 9 provinsi. Penelitian ini menegaskan keunggulan Density Based Clustering dibandingkan Partitional Based Clustering dalam data penyakit menular langsung dalam mengelompokkan provinsi ke dalam kelompok yang bermakna untuk strategi kesehatan masyarakat. Identifikasi cluster penyakit menular langsung dapat membantu otoritas kesehatan dalam menyesuaikan intervensi dan mengalokasikan sumber daya secara lebih efisien.

Kata kunci: Pengelompokan berbasis partisi, Pengelompokan berbasis densitas, Penyakit menular langsung, CLARA, DBSCAN, OPTICS.

The digital era has caused the world to produce increasingly large data, requiring more sophisticated data processing. Evaluation of clustering methods is important to obtain the optimal method for processing data. Infectious disease data in each province has a different pattern. The clustering of provinces based on the characteristics of contagious diseases is carried out so that the prevention strategy can be tailored to the needs of each province. This study aims to evaluate the effectiveness of Partitional Based Clustering and Density Based Clustering methods in grouping Indonesian provinces based on direct infectious diseases to facilitate health interventions for each province. The study used direct contagious disease data from Indonesia's health profile in 2022. Comparative analysis was conducted using the Partitional Based Clustering method (K-Means and CLARA) and Density Based Clustering method (DBSCAN and OPTICS). Optimal cluster selection in Partitional Based Clustering was determined using Gap Statistics, Elbow, and Silhouette methods. The best MinPts and epsilon were selected using the KNN method. The optimal approach was determined based on clustering evaluation methods are Calinski Harabasz index, Sum of Square Between-cluster, and C index. The Density Based Clustering approach produced the same value for both methods, with the largest Calinski Harabasz index value at 3,624.245, the largest Sum of Square Between-cluster index value at 72.131, and the smallest C index value at 0.002, indicating an optimal value. The optimal clustering grouped provinces into three clusters, Cluster 1 consists of 23 provinces, Cluster 2 consists of 2 provinces, and Cluster 3 (noise) consists of 9 provinces. This study confirms the superiority of Density Based Clustering over Partitional Based Clustering in clustering provinces based on direct infectious disease data, providing meaningful groups for public health strategies. Identifying direct infectious disease clusters can assist health authorities in tailoring interventions and allocating resources more efficiently. Keywords: Partitional-based Clustering, Density-based Clustering, Infectious diseases, CLARA, DBSCAN, OPTICS.