Random Forest Algorithm to Improve the Classification Model for Recipients of Direct Cash Assistance in Danalampah Village, Pancalang District, Kuningan Regency

Abstract

Data classification plays a crucial role in machine learning, particularly in supporting data-driven decision-making processes. Data imbalance between majority and minority classes often leads to model bias toward the majority class, reducing accuracy in detecting the minority class. In this study, recipients of direct cash assistance (BLT) represent the minority class, whose numbers are significantly fewer than non-recipients. Therefore, specific strategies are required to ensure more accurate and fair classification results. This study aims to analyze the impact of data imbalance on the performance of the BLT classification model, evaluate the model's performance using accuracy, precision, and recall metrics, and explore the effectiveness of data balancing techniques in improving the model's sensitivity to the minority class. The Random Forest algorithm serves as the primary method for building the classification model. This algorithm works by constructing numerous decision trees based on randomly selected data and features, then combining predictions through majority voting. Additionally, data balancing is implemented through manual reduction in the majority class, resulting in a dataset with a more proportional ratio (1:2), comprising 66 records: 22 BLT recipients and 44 non-recipients. The results demonstrate that the Random Forest algorithm applied to the original dataset achieves high accuracy (99.59%), but the recall for BLT recipients is 0%, indicating bias toward the majority class. After data balancing, accuracy slightly decreases to 98.33%, while recall significantly improves to 66.67%, and precision reaches 65.83%. These findings indicate that data balancing successfully enhances the model's sensitivity to the minority class without significantly compromising accuracy. This study concludes that the integration of data balancing techniques and the Random Forest algorithm improves the fairness and representation of classification outcomes. These findings offer practical contributions to the application of social data analysis, particularly in classifying social assistance recipients, to support more targeted decision-making processes