Comprehensive Evaluation of Bankruptcy Prediction in Taiwanese Firms Using Multiple Machine Learning Models
Bankruptcy prediction is a significant issue in finance because accurate predictions would enable stakeholders to act quickly to reduce their financial losses. This study developed an advanced bankruptcy prediction model using Support Vector Machines (SVM), Random Forest (RF), and Artificial Neural...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universitas Indonesia
2025-01-01
|
| Series: | International Journal of Technology |
| Subjects: | |
| Online Access: | https://ijtech.eng.ui.ac.id/article/view/7227 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Bankruptcy prediction is a significant issue in finance because accurate predictions would enable stakeholders to act quickly to reduce their financial losses. This study developed an advanced bankruptcy prediction model using Support Vector Machines (SVM), Random Forest (RF), and Artificial Neural Network (ANN) algorithms based on datasets from the UCI machine learning repository. The core contribution of this research is the establishment of a hybrid model that effectively combines multiple machine learning (ML) algorithms with advanced data with the Synthetic minority oversampling technique Tomek (SMOTE Tomek) or SMOTE- Edited Nearest Neighbor (SMOTE-ENN) resampling data technique to improve bankruptcy prediction accuracy. Additionally, a wrapper-based feature selection (FS) utilizing Binary Particle Swarm Optimization (BPSO) was utilized to find an optimal feature subset and boost the model’s predictive performance. After selecting the best features, these were used to train the three ML algorithms, and hyper-parameter optimization was implemented to boost model performance. From the results measured by evaluation metrics, the proposed model ANN with the combination of parameter tuning, feature selection algorithm, SMOTE-ENN, and optimal hyper-parameters demonstrates superior performance compared to traditional methods, achieving an F1 Score of 98.5% and an accuracy of 98.6%. The results suggest that the predictive performance of bankruptcy models can be significantly enhanced by integrating multiple analytical methodologies. This approach not only improves the accuracy but also the reliability of financial risk assessments, providing valuable insights for investors, financial analysts, and policymakers. The success of the model opens avenues for further research into hybrid predictive models in various sectors of finance, potentially transforming risk assessment methodologies. |
|---|---|
| ISSN: | 2086-9614 2087-2100 |