Penerapan Random Oversampling dan Principal Component Analysis untuk Meningkatkan Akurasi Prediksi Kebangkrutan Perusahaan di Indonesia dengan Model Machine Learning
Prediksi kebangkrutan menjadi penting untuk memberikan peringatan dini bagi manajemen dan pemangku kepentingan agar dapat mengambil tindakan preventif. Penelitian ini menguji penerapan metode Random Oversampling dan Principal Component Analysis (PCA) dalam model machine learning untuk meningkatkan akurasi prediksi kebangkrutan perusahaan. Penelitian ini menggunakan dua dataset yaitu data Taiwanese Bankruptcy Prediction dari UCI Machine Learning Repository sebanyak 6.891 data dan data primer berupa data kebangkrutan perusahaan Indonesia dari Bursa Efek Indonesia (BEI) dari tahun 2021-2023 sebanyak 2.703 data. Total keseluruhan dataset yang digunakan sebanyak 9.594 data. Empat algoritma klasifikasi—KNN, Naïve Bayes, SVM, dan Decision Tree—diuji sebelum dan sesudah penerapan metode tersebut. Hasil menunjukkan bahwa kombinasi PCA dan Random Oversampling meningkatkan recall kelas minoritas (kebangkrutan) secara signifikan. SVM menjadi algoritma terbaik dengan precision 0,86, recall 0,76, dan F1-score 0,80, sementara Decision Tree mengalami overfitting setelah oversampling. PCA berhasil mereduksi dimensi dataset hingga 98,87% varian tetap terjaga, dan Random Oversampling menyeimbangkan distribusi kelas. Abstract Bankruptcy prediction is crucial for providing early warnings to management and stakeholders to take preventive actions. This study examines the application of Random Oversampling and Principal Component Analysis (PCA) in machine learning models to improve the accuracy of corporate bankruptcy prediction. The study uses two datasets: the Taiwanese Bankruptcy Prediction data from the UCI Machine Learning Repository (6,891 data points) and primary data on Indonesian company bankruptcies from the Indonesia Stock Exchange (IDX) for 2021–2023 (2,703 data points), totaling 9,594 data points. Four classification algorithms—K-Nearest Neighbors (KNN), Naïve Bayes, Support Vector Machine (SVM), and Decision Tree—were tested before and after applying these methods. The results show that the combination of PCA and Random Oversampling significantly improved the recall of the minority class (bankruptcy). SVM emerged as the best-performing algorithm with a precision of 0.86, recall of 0.76, and F1-score of 0.80, while the Decision Tree experienced overfitting after oversampling. PCA successfully reduced the dataset’s dimensions while retaining 98.87% of the variance, and Random Oversampling balanced the class distribution.
English
Prediksi kebangkrutan menjadi penting untuk memberikan peringatan dini bagi manajemen dan pemangku kepentingan agar dapat mengambil tindakan preventif. Penelitian ini menguji penerapan metode Random Oversampling dan Principal Component Analysis (PCA) dalam model machine learning untuk meningkatkan akurasi prediksi kebangkrutan perusahaan. Penelitian ini menggunakan dua dataset yaitu data Taiwanese Bankruptcy Prediction dari UCI Machine Learning Repository sebanyak 6.891 data dan data primer berupa data kebangkrutan perusahaan Indonesia dari Bursa Efek Indonesia (BEI) dari tahun 2021-2023 sebanyak 2.703 data. Total keseluruhan dataset yang digunakan sebanyak 9.594 data. Empat algoritma klasifikasi—KNN, Naïve Bayes, SVM, dan Decision Tree—diuji sebelum dan sesudah penerapan metode tersebut. Hasil menunjukkan bahwa kombinasi PCA dan Random Oversampling meningkatkan recall kelas minoritas (kebangkrutan) secara signifikan. SVM menjadi algoritma terbaik dengan precision 0,86, recall 0,76, dan F1-score 0,80, sementara Decision Tree mengalami overfitting setelah oversampling. PCA berhasil mereduksi dimensi dataset hingga 98,87% varian tetap terjaga, dan Random Oversampling menyeimbangkan distribusi kelas. Abstract Bankruptcy prediction is crucial for providing early warnings to management and stakeholders to take preventive actions. This study examines the application of Random Oversampling and Principal Component Analysis (PCA) in machine learning models to improve the accuracy of corporate bankruptcy prediction. The study uses two datasets: the Taiwanese Bankruptcy Prediction data from the UCI Machine Learning Repository (6,891 data points) and primary data on Indonesian company bankruptcies from the Indonesia Stock Exchange (IDX) for 2021–2023 (2,703 data points), totaling 9,594 data points. Four classification algorithms—K-Nearest Neighbors (KNN), Naïve Bayes, Support Vector Machine (SVM), and Decision Tree—were tested before and after applying these methods. The results show that the combination of PCA and Random Oversampling significantly improved the recall of the minority class (bankruptcy). SVM emerged as the best-performing algorithm with a precision of 0.86, recall of 0.76, and F1-score of 0.80, while the Decision Tree experienced overfitting after oversampling. PCA successfully reduced the dataset’s dimensions while retaining 98.87% of the variance, and Random Oversampling balanced the class distribution.
Affiliations come from OpenAlex; journal OAI-PMH metadata carries none. Authors without an OpenAlex ID are matched by name, which can merge or split people.