<![CDATA[Social assistance distribution in Indonesia still faces various challenges, including inaccurate recipient data, complex bureaucracy, and lack of transparency, often resulting in misdirected aid. This study aims to optimize the application of the K-Nearest Neighbors (K-NN) method for classifying the eligibility of social assistance recipients by testing several data train-test split ratios and variations of the parameter k. The primary objective is to develop an accurate and reliable classification model to support policy-making in social assistance distribution at Kuto Batu Village, Palembang. The dataset includes citizens' socioeconomic attributes and undergoes preprocessing steps such as data cleaning, encoding, and handling missing values before being applied to the K-NN algorithm. Four data split scenarios are tested—80/20, 70/30, 60/40, and 50/50—to determine the optimal configuration. Evaluation results show model accuracies of 97.44%, 98.30%, 97.10%, and 98.00% for the respective splits. The 70/30 split yields the best performance with 98.30% accuracy, 100% precision, 98% recall, and 98.98% F1-score. This ratio is selected as the optimal configuration due to its balance between sufficient training data for pattern learning and adequate test data for evaluating model generalization. These findings demonstrate that the K-NN method is effective in objectively distinguishing eligible and ineligible recipients and has strong potential as the foundation for a decision support system to improve transparency and targeting accuracy in social assistance programs.]]>
Copyrights © 2025
