Prediksi Potensi Temuan Pemeriksaan Laporan Keuangan Pemerintah Daerah Di Regional X Menggunakan Deep Learning Berbasis Rasio Keuangan Dan Non Keuangan

Setiawan, Fery Yohan (2026) Prediksi Potensi Temuan Pemeriksaan Laporan Keuangan Pemerintah Daerah Di Regional X Menggunakan Deep Learning Berbasis Rasio Keuangan Dan Non Keuangan. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241099-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Pelaksanaan pemeriksaan Laporan Keuangan (LK) Pemerintah Daerah oleh Badan Pemeriksa Keuangan Republik Indonesia (BPK RI), khususnya pada Regional X, sering kali dihadapkan pada berbagai keterbatasan, terutama dalam hal jangka waktu pemeriksaan. Atas kondisi tersebut, umumnya untuk penentuan sampel, auditor menggunakan mekanisme uji petik. Namun hingga saat ini, setiap perwakilan BPK RI di Regional X belum memiliki suatu tool yang dapat mendukung penentuan sampel uji petik secara akurat. Penelitian sebelumnya telah melakukan analisis prediksi temuan atas hasil pemeriksaan LK menggunakan pendekatan klasifikasi multi-label, namun hanya berfokus pada temuan pemeriksaan dengan opini audit non-WTP, dan terbatas pada fitur rasio keuangan. Selain itu, pada klasifikasi multi-label, seringkali dihadapkan permasalahan ketidakseimbangan label. Oleh karena itu, pada penelitian ini, menerapkan klasifikasi multi-label untuk memprediksi seluruh potensi temuan pemeriksaan, menambahkan fitur non keuangan, dan menggunakan beberapa pendekatan resampling. Berdasarkan hasil pemodelan, metode Deep Neural Network (DNN) dengan resampling data menggunakan MLSMOTE menghasilkan performa tertinggi dibandingkan model lainnya, dengan nilai Recall sebesar 0.9298, F1 Score sebesar 0.8299, Hamming Loss sebesar 0.2380, dan Jaccard Accuracy sebesar 0.7320. Sementara itu, nilai Pre￾cision sebesar 0.7595 sedikit lebih rendah dibandingkan model lainnya. Terkait permasalahan ketidakseimbangan label, model ini juga mampu meningkatkan performa terutama pada label minoritas yang sebelumnya sulit diprediksi. Dari empat label minoritas, yaitu Y5, Y6, Y7, dan Y11, model ini berhasil meningkatkan nilai F1 Score pada label Y5, Y6, dan Y11. Dengan diperolehnya model tersebut, diharapkan hasil penelitian ini dapat dimanfaatkan sebagai tool oleh BPK RI, khususnya di Regional X, untuk mendukung pengambilan keputusan dalam penentuan sampel pada saat perencanaan pemeriksaan LK.
========================================================================================================================================
The implementation of audits of local government financial reports by the Audit Board of the Republic of Indonesia especially in Region X, often faced with various limitations, in terms of the duration of the examination. Given these conditions, examiners generally use a sampling mechanism to determine samples. However, to date, no representative of the BPK RI in Region X has a tool that can support the accurate determination of non-statistical sampling. Previous studies have analyzed predictions of findings based on financial statement audit results using a multi-label classification approach, but have only focused the potential of audit findings with non-unqualified audit opinions and been limited to financial ratio features. In addition, multi-label classification often faces the problem of label imbalance. Therefore, this study applying multi-label classification to predict all categories of audit findings, adding non-financial features, and using several resampling approaches. Based on the modeling results, the Deep Neural Network (DNN) method with data resampling using MLSMOTE produced the highest performance compared to other models, with a Recall value of 0.9298, an F1 Score of 0.8299, a Hamming Loss of 0.2380, and a Jaccard Accuracy of 0.7320. Meanwhile, the Precision value of 0.7595 is slightly lower than other models. Regarding the issue of label imbalance, this model is also able to improve performance, especially on minority labels that were previously difficult to predict. Of the four minority labels, namely Y5, Y6, Y7 and Y11, this model succeeded in increasing the F1 Score value on the Y5, Y6 and Y11 labels. With the acquisition of this model, it is hoped that the results of this study can be utilized as tool by the BPK RI, particularly in Region X, to support decision-making in sample determination when planning financial statement audits.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Learning, Klasifikasi Multi-label, Laporan Keuangan, Non Keuangan, Rasio Keuangan, Deep Learning, Multi-label Classification, Financial Statement, Non Financial, Ratio Financial
Subjects:	H Social Sciences > HB Economic Theory Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Fery Yohan Setiawan
Date Deposited:	15 Jan 2026 05:37
Last Modified:	15 Jan 2026 05:37
URI:	http://repository.its.ac.id/id/eprint/129646

Actions (login required)

View Item