DETEKSI FINANCIAL STATEMENT FRAUD MENGGUNAKAN RANDOM FOREST PADA PERUSAHAAN MANUFAKTUR YANG TERDAFTAR DI BURSA EFEK INDONESIA (BEI)
Kecurangan laporan keuangan (Financial Statement Fraud) merupakan masalah serius yang dapat merugikan investor dan pengguna laporan keuangan. Seiring perkembangan teknologi, metode konvensional dinilai belum optimal sehingga diperlukan pendekatan Machine Learning. Penelitian ini bertujuan untuk mengetahui kemampuan Random Forest dalam mendeteksi Financial Statement Fraud serta mengidentifikasi variabel yang paling berkontribusi berdasarkan Fraud Triangle. Penelitian ini menggunakan pendekatan kuantitatif pada perusahaan manufaktur yang terdaftar di Bursa Efek Indonesia (BEI) periode 2021–2024. Sampel penelitian terdiri atas 73 perusahaan dengan 292 observasi yang dipilih menggunakan teknik purposive sampling. Deteksi fraud dilakukan menggunakan Beneish M-Score dengan variabel Financial Target, External Pressure, Nature of Industry, dan Change in Auditor. Data diolah melalui pre-processing, SMOTE, dan pemodelan Random Forest. Hasil penelitian menunjukkan bahwa Random Forest mampu mendeteksi fraud dengan nilai accuracy sebesar 0,7627. Variabel yang paling dominan adalah Nature of Industry, diikuti Financial Target dan External Pressure, sedangkan Change in Auditor memiliki kontribusi paling kecil. Kesimpulannya, Random Forest dapat digunakan sebagai metode dalam mendeteksi Financial Statement Fraud, meskipun masih perlu pengembangan lebih lanjut. Financial statement fraud is a serious issue that can harm investors and users of financial statements. By advancement of technology, conventional methods are considered suboptimal, necessitating a Machine Learning approach. This study aims to determine the ability of Random Forest to detect financial statement fraud and to identify the variables that contribute most significantly based on the Fraud Triangle. This study employed a quantitative approach on manufacturing companies listed on the Indonesia Stock Exchange (IDX) for the period 2021–2024. The research sample consisted of 73 companies with 292 observations selected using purposive sampling. Fraud detection was performed using the Beneish M-Score with the variables Financial Target, External Pressure, Nature of Industry, and Change in Auditor. The data was processed through pre-processing, SMOTE, and Random Forest modeling. The results showed that Random Forest is capable of detecting fraud with an accuracy of 0.7627. The most dominant variable is Nature of Industry, followed by Financial Target and External Pressure, while Change in Auditor has the smallest contribution. In conclusion, Random Forest can be used as a method for detecting Financial Statement Fraud, although further development is still needed.