PRAMUDYA, Reza Iqbal and Sugiharto, Aris and Somantri, Maman (2026) HYBRID GENETIC ALGORITHM DAN PARTICLE SWARM OPTIMIZATION SEBAGAI INISIALISASI CENTROID PADA K-MEANS CLUSTERING UNTUK DATASET GIZI NASIONAL. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text cover aja.pdf Download (116kB)
	Text COVER OK.pdf Restricted to Repository staff only Download (10MB)
	Text BAB I.pdf Download (248kB)
	Text BAB II.pdf Download (672kB)
	Text BAB III.pdf Restricted to Repository staff only Download (1MB)
	Text BAB IV.pdf Restricted to Repository staff only Download (1MB)
	Text BAB V.pdf Restricted to Repository staff only Download (239kB)
	Text DAFTAR PUSTAKA.pdf Download (264kB)
	Text LAMPIRAN.pdf Restricted to Repository staff only Download (373kB)

Abstract

Kesenjangan status gizi antarwilayah di Indonesia menuntut metode pemetaan komputasional yang presisi guna mendukung efektivitas kebijakan publik. Algoritma K-Means konvensional rentan terhadap jebakan local optima akibat inisialisasi centroid acak. Penelitian ini mengusulkan arsitektur hibrida Genetic Algorithm–Particle Swarm Optimization (GAPSO) untuk mengoptimalkan klasterisasi 514 kabupaten dan kota berdasarkan data gizi nasional tahun 2023. Prapemrosesan data mengimplementasikan Winsorization bertahap dan normalisasi Min-Max Scaling guna mengendalikan pencilan ekstrem sekaligus menyeragamkan skala fitur. Evaluasi mengeksplorasi variasi K=2 hingga K=10 dengan probabilitas penyempurnaan 0,1 hingga 0,9. Konfigurasi K=3 ditetapkan sebagai arsitektur optimal dengan K-Means+GAPSO probabilitas 0,1; 0,6; dan 0,8 mencatatkan Silhouette Score terbaik sebesar 0,2948 dan Davies-Bouldin Index terendah sebesar 1,2236. Analisis Permutation Importance mengidentifikasi Prevalence of Undernourishment (PoU), Skor PPH, Konsumsi Protein, dan Konsumsi Energi sebagai variabel pembeda paling dominan. Pemetaan menghasilkan tiga tipologi: Klaster 0 (131 wilayah, sangat rentan gizi, rata-rata PoU 23,23%), Klaster 2 (264 wilayah, rentan sedang), dan Klaster 1 (119 wilayah, tahan gizi). Temuan ini menyumbangkan landasan analitik bagi pemerintah dalam merancang intervensi gizi berbasis kerentanan wilayah yang lebih terarah dan presisi.
Kata Kunci: Klasterisasi, K-Means, Genetic Algorithm, Particle Swarm Optimization, Ketahanan Gizi

Nutritional status disparities across Indonesian regions demand precise computational mapping methods to support effective public policy. Conventional K-Means algorithms are prone to local optima entrapment due to random centroid initialization. This study proposes a Genetic Algorithm–Particle Swarm Optimization (GAPSO) hybrid architecture to optimize the clustering of 514 districts and cities based on 2023 national nutritional data. Data preprocessing implements iterative Winsorization combined with Min-Max Scaling normalization to control extreme outliers while standardizing feature scales. Evaluations explored K=2 through K=10 with refinement probabilities ranging from 0.1 to 0.9. The K=3 configuration was established as the optimal architecture, with K-Means+GAPSO at refinement probabilities 0.1, 0.6, and 0.8 consistently achieving the best Silhouette Score of 0.2948 and the lowest Davies-Bouldin Index of 1.2236. Permutation Importance analysis identified Prevalence of Undernourishment (PoU), PPH Score, Protein Consumption, and Energy Consumption as the most dominant discriminating variables. The mapping produced three typologies: Cluster 0 (131 regions, highly nutritionally vulnerable, mean PoU 23.23%), Cluster 2 (264 regions, moderately vulnerable), and Cluster 1 (119 regions, nutritionally resilient). These findings provide an analytical foundation for government authorities to design more targeted and precise nutrition interventions based on regional vulnerability.
Keywords: Clustering, K-Means, Genetic Algorithm, Particle Swarm Optimization, Nutritional Security

Actions (login required)

View Item