Dinda Pusparahmi Sholawatunnisa, Heru Supriyanto

COMPARATIVE FORECASTING EARLY WARNING SYSTEM DEVIASI ANGGARAN PEMERINTAH BERBASIS MACHINE LEARNING: STUDI EMPIRIS IKPA KANWIL DJPB PROVINSI NTB SEBAGAI BUN

Introduction

Comparative forecasting early warning system deviasi anggaran pemerintah berbasis machine learning: studi empiris ikpa kanwil djpb provinsi ntb sebagai bun. Peringatan dini deviasi anggaran pemerintah berbasis ML membandingkan SARIMA, XGBoost, Random Forest pada IKPA DJPB NTB. SARIMA paling akurat, dukung perencanaan & akuntabilitas APBN.

Abstract

Transformasi digital treasury dalam pengelolaan keuangan negara memerlukan pendekatan berbasis data untuk meningkatkan akuntabilitas APBN. Indikator Kinerja Pelaksanaan Anggaran (IKPA) menjadi instrumen utama dalam mengukur kinerja anggaran pada satuan kerja. Namun, dari delapan indikator penyusun IKPA, selama tiga tahun terakhir, komponen deviasi halaman III DIPA menunjukkan nilai terendah, mengindikasikan adanya tantangan dalam akurasi perencanaan dan eksekusi anggaran. Penelitian ini membandingkan tiga model prediktif berbasis artificial intelligence untuk meramalkan nilai indikator deviasi halaman III DIPA: SARIMA (Seasonal Autoregressive Integrated Moving Average) yang handal dalam menangkap pola musiman, XGBoost (Extreme Gradient Boosting) yang unggul dalam optimasi gradien, dan Random Forest yang kuat dalam menangani kompleksitas data. Dataset mencakup observasi bulanan nilai indikator deviasi halaman III DIPA dari Januari 2022 sampai dengan September 2024, memberikan dasar yang komprehensif untuk analisis time series. Metodologi penelitian menerapkan pendekatan kuantitatif dengan preprocessing data, pemilihan fitur, dan validasi silang untuk memastikan robustness model. Evaluasi performa menggunakan metrik MAE, RMSE, dan MAPE. Temuan penelitian mengungkapkan bahwa Model SARIMA memberikan akurasi prediksi tertinggi dengan error rate terendah. Kontribusi penelitian ini signifikan dalam dua aspek: pengembangan early warning system untuk deviasi anggaran dan penyediaan tools pendukung keputusan berbasis AI untuk perencanaan anggaran yang lebih akurat. Implementasi model ini diharapkan dapat membantu satuan kerja meningkatkan nilai IKPA mereka secara keseluruhan.

Review

The paper, "COMPARATIVE FORECASTING EARLY WARNING SYSTEM DEVIASI ANGGARAN PEMERINTAH BERBASIS MACHINE LEARNING: STUDI EMPIRIS IKPA KANWIL DJPB PROVINSI NTB SEBAGAI BUN," addresses a critical issue in digital treasury management: improving budget accountability through data-driven approaches. It identifies a persistent challenge within the Indicator Kinerja Pelaksanaan Anggaran (IKPA), specifically the "deviasi halaman III DIPA" component, which has consistently shown the lowest performance over three years. This low performance signals fundamental issues in budget planning and execution accuracy, highlighting the urgent need for tools to proactively manage these deviations. The research aims to develop and compare advanced machine learning models to forecast this critical indicator, ultimately proposing an an early warning system to enhance the precision of government budget management. To achieve its objective, the study employs a rigorous quantitative methodology, comparing three distinct artificial intelligence models: SARIMA (Seasonal Autoregressive Integrated Moving Average), known for its ability to capture seasonal patterns; XGBoost (Extreme Gradient Boosting), recognized for its optimization capabilities; and Random Forest, valued for handling complex datasets. The models are trained and evaluated on a comprehensive monthly dataset of the "deviasi halaman III DIPA" indicator, spanning from January 2022 to September 2024. The methodological approach includes thorough data preprocessing, strategic feature selection, and robust cross-validation techniques to ensure the reliability and generalizability of the models. Performance evaluation is conducted using standard metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE), providing a holistic assessment of prediction accuracy. The empirical findings reveal that among the compared models, SARIMA demonstrates superior performance, yielding the highest prediction accuracy with the lowest error rates. This outcome underscores the significant seasonal patterns inherent in budget deviation data and SARIMA's effectiveness in capturing them. The research makes substantial contributions by not only developing a practical early warning system for government budget deviations but also by providing an AI-based decision support tool for more accurate budget planning. The successful implementation of this model is anticipated to empower *satuan kerja* (work units) in the NTB province to proactively manage and improve their IKPA values, thereby fostering greater accountability and efficiency in public finance management. This study lays a strong foundation for integrating advanced analytics into national treasury operations.

Full Text

Comments

You need to be logged in to post a comment.