Klasifikasi Status Kelolosan Calon Penerima Bantuan Program Keluarga Harapan Berdasarkan DTKS Desa Menggunakan Random Forest

Abstract

Poverty remains a challenge for the targeted distribution of social assistance, as discrepancies between the poor population and registered beneficiary families risk causing inclusion and exclusion errors. The Integrated Social Welfare Data (DTKS) is a critical basis for determining beneficiaries of the Family Hope Program (PKH), though its use remains largely administrative. In Rowo Indah Village, 2,192 of 2,449 residents registered in the DTKS had not been recorded as PKH recipients. This study aimed to implement a Random Forest algorithm to classify the eligibility status of prospective PKH recipients using 2025 DTKS data. A quantitative approach was employed, involving preprocessing, feature engineering, Spearman and Chi-square correlation analysis, Random Forest modeling, and confusion matrix evaluation. Aggregation yielded 901 unique household records, hyperparameters were optimized via GridSearchCV and 5-fold Stratified K-Fold cross-validation, and a balanced class_weight parameter was applied to address class imbalance. The 60:40 split scenario delivered the best performance, achieving 76% accuracy, 72% F1-macro, 56% precision, and 67% recall. Household size, number of school-aged children, number of dependents, number of toddlers, and presence of a pregnant mother were significantly correlated with PKH participation status. These findings indicate that Random Forest effectively captures household-level patterns associated with PKH eligibility, with satisfactory recall for the minority class despite room for precision improvement. The resulting model can serve as a preliminary verification tool for village governments, though field validation remains necessary for accurate targeting