Electronic Theses and Dissertation

Kinerja anggaran merupakan indikator dalam menilai efektivitas pengelolaan anggaran daerah, khususnya pada Organisasi Perangkat Daerah (OPD). Rendah dan tingginya tingkat kinerja anggaran atau tidak tercapainya realisasi anggaran dapat mencerminkan permasalahan dalam perencanaan, penganggaran, maupun pelaksanaan program. Oleh karena itu, diperlukan analisis determinan kinerja anggaran kabupaten/kota di Provinsi Aceh dengan pemanfaatan machine learning dengan menggunakan 254 observasi OPD tahun 2023 dan empat variabel, yaitu varian anggaran tahun lalu, besaran anggaran, perubahan anggaran, dan jenis OPD. Penelitian ini bertujuan untuk membandingkan kinerja tiga model ensemble regression, yaitu Random Forest Regression (RFR), XGBoost Regression (XGBR), dan CatBoost Regression (CBR) dalam memprediksi dan mengidentifikasi variabel determinan kinerja anggaran, evaluasi model terbaik menggunakan MSE, RMSE, MAE dan R². Namun, keterbatasan interpretabilitas pada model machine learning diatasi dengan menggunakan SHapley Additive exPlanations (SHAP). Hasil penelitian menunjukkan model bahwa model CatBoost Regression (CBR) memberikan performa kinerja prediksi terbaik (MSE sebesar 〖1,4 ×10〗^(-5), RMSE sebesar〖3,694 ×10〗^(-3), MAE sebesar〖2,708 ×10〗^(-3), R² sebesar 0,5900 (59%), dan Adjusted-R2 sebesar 0,5544 (55,44%)). Sementara, interpretabilitas menujukkan varian anggaran tahun lalu merupakan variabel paling berkontribusi dan berpengaruh, diikuti perubahan anggaran, besaran anggaran, dan jenis OPD terhadap kinerja anggaran OPD.

Budget performance is an important indicator in assessing the effectiveness of regional budget management, particularly within Local Government Organizations (OPD). Low or high levels of budget performance, as well as the failure to achieve budget realization, may reflect issues in planning, budgeting, and program implementation. Therefore, this study analyzes the determinants of budget performance across districts/cities in Aceh Province using a machine learning approach, based on 254 OPD observations in 2023 and four variables: prior year budget variance, budget size, budget changes, and OPD type. This study aims to compare the performance of three ensemble regression models, namely Random Forest Regression (RFR), XGBoost Regression (XGBR), and CatBoost Regression (CBR), in predicting and identifying the determinants of budget performance. Model evaluation is conducted using MSE, RMSE, MAE, and R² metrics. However, due to the limited interpretability of machine learning models, SHapley Additive exPlanations (SHAP) are employed to enhance model interpretability. The results indicate that CatBoost Regression (CBR) provides the best predictive performance, with an (MSE of 〖1,4 ×10〗^(-5), RMSE of 〖3,694 ×10〗^(-3), MAE of 〖2,708 ×10〗^(-3), R² of 0.5900 (59%), and Adjusted-R2 of 0,5544 (55,44%)). Furthermore, SHAP interpretability showing that prior year budget variance is the most influential variable, followed by budget changes, budget size, and OPD type in determining OPD budget performance.