Stunting masih menjadi masalah kesehatan masyarakat di Indonesia dengan prevalensi 19,8% pada tahun 2024. Penetapan prioritas intervensi umumnya dilakukan pada tingkat populasi sehingga belum mampu mengakomodasi perbedaan profil risiko individu. Penelitian ini bertujuan mengembangkan kerangka penetapan prioritas intervensi penurunan stunting berbasis profil risiko individu menggunakan pendekatan machine learning. Penelitian menggunakan desain potong lintang dengan data sekunder Survei Kesehatan Indonesia (SKI) 2023 dan Survei Status Gizi Indonesia (SSGI) 2024 pada 381.108 balita usia 0–59 bulan. Analisis dilakukan pada tiga subgrup usia (0–5 bulan, 6–23 bulan, dan 24–59 bulan) menggunakan algoritma Logistic Regression, Random Forest, dan XGBoost. Kinerja model dievaluasi berdasarkan ROC-AUC, precision, recall, dan F1-score. Hasil menunjukkan bahwa Logistic Regression merupakan model terbaik pada usia 0–5 bulan (ROC-AUC=0,764), sedangkan XGBoost Tuned menjadi model terbaik pada usia 6–23 bulan (ROC-AUC=0,663) dan 24–59 bulan (ROC-AUC=0,661). Faktor prediktor dominan serta prioritas intervensi berbeda pada setiap subgrup usia. Implementasi model melalui dashboard menghasilkan rekomendasi prioritas intervensi berdasarkan perubahan probabilitas risiko setelah faktor intervensi dimodifikasi pada profil individu. Pendekatan ini menghasilkan kerangka penetapan prioritas intervensi berbasis risiko individu yang berpotensi mendukung pengambilan keputusan penurunan stunting secara lebih presisi, preventif, dan berbasis data.

Stunting remains a major public health problem in Indonesia, with a prevalence of 19.8% in 2024. Current intervention prioritization is generally conducted at the population level and does not adequately account for individual risk profiles. This study aimed to develop an individual risk-based framework for prioritizing stunting interventions using machine learning approaches. A cross-sectional study was conducted using secondary data from the 2023 Indonesia Health Survey (SKI) and the 2024 Indonesia Nutritional Status Survey (SSGI), involving 381,108 children aged 0–59 months. Analyses were performed across three age subgroups (0–5 months, 6–23 months, and 24–59 months) using Logistic Regression, Random Forest, and XGBoost algorithms. Model performance was evaluated using ROC-AUC, precision, recall, and F1-score. The results showed that Logistic Regression achieved the best performance in the 0–5 months subgroup (ROC-AUC = 0.764), while XGBoost Tuned performed best in the 6–23 months (ROC-AUC = 0.663) and 24–59 months (ROC-AUC = 0.661) subgroups. The dominant predictor factors and intervention priorities differ for each age subgroup. Implementing the model through a dashboard generates intervention priority recommendations based on changes in risk probability after intervention factors are adjusted for an individual's profile. This approach creates a risk-based intervention priority framework that could support more precise, preventive, and data-driven decisions for reducing stunting.

Latar Belakang: Unmet need keluarga berencana (KB) masih menjadi tantangan kesehatan masyarakat di Indonesia. Intervensi yang tersedia umumnya dirancang secara seragam sehingga belum mempertimbangkan keragaman faktor perilaku dan psikososial yang melatarbelakangi kebutuhan KB yang tidak terpenuhi. Pendekatan segmentasi populasi dapat membantu mengidentifikasi kelompok perempuan dengan karakteristik yang berbeda sehingga program KB dapat dirancang lebih tepat sasaran. Metode: Penelitian ini menggunakan data sekunder Community Based Survey (CBS) Guttmacher tahun 2018 pada 1.430 perempuan usia 15–49 tahun dengan unmet need KB di enam provinsi di Pulau Jawa. Analisis cluster dilakukan menggunakan Gower Distance dengan metode HDBSCAN dan K-Means, dengan validasi menggunakan average silhouette, PERMANOVA, Betadisper, dan uji Chi-Square/Fisher. Hasil: HDBSCAN menghasilkan tiga cluster (206 responden) dengan 1.224 responden (85,6%) sebagai noise, sedangkan K-Means mengelompokkan seluruh 1.430 responden ke dalam tiga cluster. Meskipun menghasilkan struktur pengelompokan yang berbeda, kedua metode mengidentifikasi tiga profil hambatan yang saling melengkapi. Kelompok pertama tidak memiliki riwayat penggunaan kontrasepsi namun memiliki otonomi reproduksi tinggi. Kelompok kedua memiliki pengalaman penggunaan kontrasepsi yang luas disertai pengalaman efek samping yang nyata. Kelompok ketiga didominasi oleh rendahnya otonomi reproduksi dan kuatnya penerimaan norma gender tradisional. Secara substantif, kedua metode secara konsisten mengidentifikasi empat pola hambatan utama, yaitu mispersepsi terhadap kontrasepsi, penghentian penggunaan akibat efek samping, minimnya pengalaman penggunaan kontrasepsi, serta rendahnya otonomi reproduksi. Riwayat penggunaan kontrasepsi, pengalaman efek samping, persepsi terhadap metode kontrasepsi, dan otonomi reproduksi merupakan faktor pembeda utama antar kelompok. Implikasi Program: Perempuan dengan unmet need KB bukan merupakan kelompok yang homogen. Segmentasi berbasis perilaku dan psikososial mampu mengidentifikasi perbedaan karakteristik yang tidak dapat dijelaskan hanya melalui faktor sosiodemografi, sehingga program KB yang mempertimbangkan profil hambatan spesifik tiap kelompok berpotensi menghasilkan intervensi yang lebih tepat sasaran dan efektif.

Background: Unmet need for family planning (FP) remains a public health challenge in Indonesia. Available interventions are generally designed using a uniform approach and have not fully accounted for the diversity of behavioral and psychosocial factors underlying unmet need for family planning. A population segmentation approach may help identify groups of women with distinct characteristics, enabling the development of more targeted family planning programs. Methods: This study used secondary data from the 2018 Guttmacher Community Based Survey (CBS) involving 1,430 women aged 15–49 years with unmet need for family planning across six provinces in Java. Cluster analysis was conducted using Gower Distance with HDBSCAN and K-Means methods, with validation performed using average silhouette, PERMANOVA, Betadisper, and Chi-Square/Fisher tests. Results: HDBSCAN identified three clusters (206 respondents) while 1,224 respondents (85.6%) were classified as noise, whereas K-Means assigned all 1,430 respondents into three clusters. Despite producing different clustering structures, both methods identified three complementary barrier profiles. The first group had no prior contraceptive use history but demonstrated high reproductive autonomy. The second group had extensive contraceptive use experience accompanied by actual side effect experiences. The third group was characterized by low reproductive autonomy and strong acceptance of traditional gender norms. Substantively, both methods consistently identified four major barrier patterns, namely contraceptive misperceptions, discontinuation due to side effects, limited contraceptive experience, and low reproductive autonomy. Contraceptive use history, side effect experiences, perceptions of contraceptive methods, and reproductive autonomy emerged as the primary factors differentiating the identified groups. Program Implications: Women with unmet need for family planning do not constitute a homogeneous population. Behavioral and psychosocial segmentation was able to identify characteristic differences that could not be adequately explained by sociodemographic factors alone, suggesting that family planning programs incorporating specific barrier profiles for each group have the potential to produce more targeted and effective interventions.

Stunting masih menjadi tantangan kesehatan masyarakat di Indonesia dengan prevalensi nasional sebesar 19,8% pada tahun 2024. Meskipun prevalensinya menurun, beban kasus masih terkonsentrasi di Pulau Sumatera dan Jawa yang menampung sebagian besar balita Indonesia. Kesenjangan karakteristik antarkabupaten/kota di kedua pulau tersebut sering kali tertutupi oleh rata-rata tingkat provinsi, sehingga intervensi yang diterapkan cenderung bersifat seragam dan kurang tepat sasaran. Penelitian ini bertujuan mengidentifikasi karakteristik wilayah berdasarkan indikator Indeks Pembangunan Manusia (IPM) dan determinan stunting, mendeskripsikan karakteristik setiap cluster, serta menyusun rekomendasi kebijakan berbasis karakteristik wilayah. Penelitian menggunakan desain studi ekologi dengan unit analisis 273 kabupaten/kota di Pulau Sumatera dan Jawa berdasarkan data SUSENAS dan Survei Status Gizi Indonesia (SSGI) Tahun 2024. Analisis dilakukan menggunakan metode K-Means Clustering terhadap empat indikator IPM dan enam determinan stunting, sedangkan prevalensi stunting digunakan sebagai variabel pembanding melalui uji ANOVA. Hasil penelitian mengidentifikasi empat cluster dengan karakteristik yang berbeda. Cluster 3 memiliki tingkat pembangunan manusia tertinggi dan prevalensi stunting terendah, sedangkan Cluster 4 memiliki prevalensi stunting tertinggi yang disertai tingkat kemiskinan lebih tinggi serta akses air minum dan sanitasi yang lebih rendah. Temuan penting menunjukkan bahwa Cluster 4 memiliki rasio puskesmas dan bidan yang relatif tinggi, namun tetap mencatat prevalensi stunting tertinggi, mengindikasikan bahwa intervensi layanan kesehatan saja belum cukup tanpa perbaikan determinan sosial dan lingkungan. Uji ANOVA menunjukkan perbedaan prevalensi stunting yang bermakna antarcluster (p<0,001). Penelitian ini menyimpulkan bahwa pendekatan kebijakan berbasis karakteristik  wilayah diperlukan agar strategi percepatan penurunan stunting dapat disesuaikan dengan karakteristik masing-masing cluster sehingga menjadi lebih tepat sasaran, efektif, dan berkelanjutan.

Stunting remains a major public health challenge in Indonesia, with a national prevalence of 19.8% in 2024. Although the prevalence has declined, the absolute burden of cases remains concentrated in the islands of Sumatra and Java, which are home to the majority of Indonesia's under-five population. Disparities among regencies and municipalities are often masked by provincial averages, resulting in uniform interventions that may not adequately address local needs. This study aimed to identify regional typologies based on the Human Development Index (HDI) indicators and stunting determinants, describe the characteristics of each cluster, and develop policy recommendations tailored to regional characteristics. An ecological study design was employed using secondary data from 273 regencies and municipalities in Sumatra and Java obtained from the 2024 National Socioeconomic Survey (SUSENAS) and the 2024 Indonesian Nutrition Status Survey (SSGI). The K-Means Clustering method was applied to four HDI indicators and six stunting determinants, while stunting prevalence was used as a validation variable through analysis of variance (ANOVA). The analysis identified four distinct regional clusters with different characteristics. Cluster 3 exhibited the highest level of human development and the lowest stunting prevalence, whereas Cluster 4 had the highest stunting prevalence, accompanied by higher poverty rates and lower access to improved drinking water and sanitation. A notable finding was that Cluster 4 had relatively high ratios of primary healthcare centers and midwives despite recording the highest stunting prevalence, indicating that healthcare interventions alone are insufficient without addressing broader social and environmental determinants. ANOVA confirmed significant differences in stunting prevalence among the clusters (p<0.001). This study concludes that a regional typology-based approach is essential for designing more targeted, effective, and sustainable stunting reduction policies that are tailored to the characteristics of each cluster.