Implementasi Algoritma DBSCAN untuk Pengelompokan Perilaku Merokok pada UK Smoking Survey Dataset

Maria Winarni Br. Silitonga (Ilmu Komputer, Fakultas Matematika dan Ilmu Pengetahuan Alam, Universitas Negeri Medan)
Juliana Gloria Br. Sipayung (Ilmu Komputer, Fakultas Matematika dan Ilmu Pengetahuan Alam, Universitas Negeri Medan)
Nazwa Salsyabilla Ramadhani (Ilmu Komputer, Fakultas Matematika dan Ilmu Pengetahuan Alam, Universitas Negeri Medan)
M. Fahmi Arafat (Ilmu Komputer, Fakultas Matematika dan Ilmu Pengetahuan Alam, Universitas Negeri Medan)

Journal Of Informatics And Busisnes
Vol. 4 No. 1 (2026): April - Juni

This study applies the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to the UK Smoking Survey Dataset from Kaggle (1,691 records, 13 attributes). Preprocessing includes missing value imputation, label encoding, feature selection of 8 features, and StandardScaler normalization. Optimal parameters (eps=2.0; min_samples=15) were determined via K-Distance Graph. Four clusters were identified: Cluster 0 (n=539, male non-smokers, avg. 51.2 yrs), Cluster 1 (n=228, female smokers, 11.9 cig/weekday), Cluster 2 (n=731, female non-smokers, avg. 53.0 yrs), Cluster 3 (n=168, male smokers, 13.5 cig/weekday), and 25 noise points as extreme heavy smokers. Evaluation: Silhouette Score=0.2032, Davies-Bouldin Index=2.0494, Calinski-Harabasz Index=395.68. Results demonstrate DBSCAN’s effectiveness in identifying demographic-based smoking behavior patterns.