Comparative study of support vector machines and random forests machine learning algorithms on credit operation

• Published in: Software, practice & experience practice & experience, 2021-12, Vol.51 (12), p.2492-2500
• Main Authors: Teles, Germanno, Rodrigues, Joel J. P. C., Rabêlo, Ricardo A. L., Kozlov, Sergei A.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Bankruptcy; Big Data; Comparative studies; credit operation; Data analysis; Data collection; Gaussian process; Independent variables; Insolvency; International finance; Machine learning; Mathematical models; Qualitative analysis; random forests; Risk management; Support vector machines
• ISSN: 0038-0644; 1097-024X
• DOI: 10.1002/spe.2842

Summary Corporate insolvency has significant adverse effects on an economy. With the number of multinationals increasing rapidly, corporate bankruptcy can severely disrupt the global financial environment. However, multinationals do not fail instantaneously; objective strategies combined with a rigorous analysis of both qualitative and quantifiable data can go a long way in identifying an organization's financial risks. Recent advancements in information and communication technologies have made data collection and storage an easy task. The challenge becomes mining the appropriate data about a company's financial risks and implementing it in forecasting a company's insolvency probabilities. In recent years, machine learning has been incorporated into big data analytics owing to its massive success in learning complex models. Machine learning algorithms such as Support Vector Machines (SVM), Random Forests (RF), Artificial Neural Networks, Gaussian Processes, and Adaptive Learning have been used in the analysis of Big Data to predict the financial risks of companies. In this paper, credit scoring is explored with regards to data processed using the collateral as an independent variable. The obtained results indicate that RF algorithm is promising for use in credit risk management. This research shows the advantages of the RF approach over the SVM algorithm are its speed and operational simplicity, and SVM has the benefit of higher classification accuracy than RF. The paper compares the SVM and RF algorithms to forecast the recovered value in a credit task. The execution of the projected intelligent systems uses tests and algorithms for authentication of the projected model.