Predicting Budget Absorption Categories Using Random Forest and Support Vector Machine Methods

Abstract

Budget classification plays a crucial role in planning, management, and budgeting, from implementation to accountability. We create budgets by considering various types of expenditures and funding sources. Each type of expenditure, such as employee salaries, goods, capital, grants, social assistance, subsidies, interest, and non-tax revenue (PNBP) or public service agencies (BLU), has its own set of rules and methods for tracking money. This study aims to demonstrate how budget classification, based on expenditure types and funding sources, is applied in the implementation of the Revenue Budget. This study aims to assess the classification performance of two models, namely the Random Forest Classifier (RFC) and Support Vector Machine (SVM), based on historical data and evaluate the performance of each model. Tests show that the Random Forest model consistently outperforms the SVM model for each data proportion, with a ratio of 90:10 to 60:40. The Random Forest model achieved its best performance at the 80:20 data split, with an accuracy score of 94 percent, a precision score of 94 percent, a recall score of 94 percent, and an F1 score of 87 percent. The average accuracy score of the SVM test results was 80 percent.