Importance The article reviews the process of forecasting the creditworthiness of the bank's customers. As competition in the lending market gains momentum, it would be reasonable to forge new components of the process and assess the credit risk more accurately. Objectives The objective of the research is to improve methods for forecasting the customers' creditworthiness by using contemporary machine learning methods and taking optimal decisions on granting loans. Methods We propose an algorithm for forecasting the creditworthiness using the customer's profile and machine learning methods (clustering, regression analysis, and classification). The algorithm enables researchers to use separate models and their possible combinations. As for the approach proposed in the article, we suggest performing a preliminary analysis of data (discretization, search for statistically significant features of the borrower) and applying various quality criteria to choose an optimal structure. Based on the results, the bank's customers are divided by the given number of classes k. Results Based on the algorithm, we generated an effective method for forecasting the creditworthiness to assess the probability of loan repayment in line with the available profile of the customer. The efficiency of this method is proved with the case study. Based on 20 features of the borrower, we built various classification models (both separately and in various combinations). We found a structure with the least mean square error of forecasting. When customers are split into more than two classes, it streamlines the process of making loan decisions since the default risk decreases. Conclusions and Relevance A combination of models and machine learning improves creditworthiness forecasts, enhances the quality of risk assessment and streamlines the lending process.
Keywords: creditworthiness, machine learning, discriminant analysis, support vectors, logistic regression
References:
Aivazyan S.A., Bukhshtaber V.M., Enyukov I.S., Meshalkin L.D. Prikladnaya statistika: Klassifikatsiya i snizhenie razmernosti [Applied statistics: Classification and reduction of dimensions]. Moscow, Finansy i statistika Publ., 1989, 607 p.
Gotovkin I. Kompleksnaya skoringovaya model' otsenki defolta klienta [A comprehensive scoring model for assessing the customer’s default]. Bankovskie tekhnologii = Banking Technologies, 2006, no. 1, pp. 27–35.
Litvinova S.A. Skoringovye sistemy kak sredstvo minimizatsii kreditnogo riska banka [Scoring systems as a tool to mitigate the credit risk of the bank]. Audit i finansovyi analiz = Audit and Financial Analysis, 2010, no. 2, pp. 396–397.
Chernyi I.M. Kreditnyi skoring: rossiiskii variant razvitiya [Credit score: the Russian version of development]. Bankovskie uslugi = Banking Services, 2006, no. 4, pp. 12–17.
Ishina I.V., Sazonova M.N. Skoring – model' otsenki kreditnogo riska [Score is a model for credit risk assessment]. Audit i finansovyi analiz = Audit and Financial Analysis, 2007, no. 4, pp. 297–304.
Dubrov A.M., Mkhitaryan V.S., Troshin L.I. Mnogomernye statisticheskie metody [Multivariate statistical methods]. Moscow, Finansy i statistika Publ., 2003, 352 p.
Sorokin A.S. Postroenie skoringovykh kart s ispol'zovaniem modeli logisticheskoi regressii [Building scorecards by using a logistic regression model]. Available at: Link. (In Russ.)
Vasil'ev N.P., Egorov A.A. Opyt rascheta parametrov logisticheskoi regressii metodom N'yutona–Rafsona dlya otsenki zimostoikosti rastenii [Experience of calculating logistic regression parameters by the Newton–Rafson method to assess cold resistance of plants]. Matematicheskaya biologiya i bioinformatika = Mathematical Biology and Bioinformatics, 2011, vol. 6, no. 2, pp. 190–199.
Wasserman Ph. Neirokomp'yuternaya tekhnika: Teoriya i praktika [Neural Computing: Theory and Practice]. Moscow, Mir Publ., 1992, 184 p.
Andreeva G.V. Skoring kak metod otsenki kreditnogo riska [Scoring as a method for assessing the credit risk]. Bankovskie tekhnologii = Banking Technologies, 2000, no. 6, pp. 14–19.
Yakupov A.I. Primenenie derev'ev reshenii dlya modelirovaniya kreditosposobnosti klientov kommercheskogo banka [The use of decision trees for modeling the creditworthiness of the commercial bank’s customers]. Iskusstvennyi intellekt = Artificial Intelligence, 2008, no. 4, pp. 208–213.
Vapnik V.N. Vosstanovlenie zavisimostei po empiricheskim dannym [Restoring the dependency based on empirical data]. Moscow, Nauka Publ., 1979, 448 p.
Aivazyan S.A., Enyukov I.S., Meshalkin L.D. Prikladnaya statistika. Osnovy modelirovaniya i pervichnaya obrabotka dannykh [Applied statistics. Fundamentals of modeling and primary processing of data]. Moscow, Finansy i statistika Publ., 1983, 471 p.
Vagin V.N., Golovina E.Yu., Zagoryanskaya A.A., Fomina M.V. Dostovernyi i pravdopodobnyi vyvod v intellektual'nykh sistemakh [Credible and plausible inference in intelligent systems]. Moscow, FIZMATLIT Publ., 2004, 704 p.
Malyugin V.I., Grin' N.V. Ob effektivnosti statisticheskikh algoritmov kreditnogo skoringa [On the effectiveness of statistical algorithms for credit score]. Банкаўскі веснік, 2010, no. 31, pp. 39–46.
Little R.J.A., Rubin D.B. Statisticheskii analiz dannykh s propuskami [Statistical Analysis with Missing Data]. Moscow, Finansy i statistika Publ., 1990, 336 p.
Valeev S.G. Regressionnoe modelirovanie pri obrabotke dannykh [Regression modeling in data processing]. Kazan, FEN Publ., 2001, 296 p.
Draper N., Smith H. Prikladnoi regressionnyi analiz. Mnozhestvennaya regressiya [Applied Regression Analysis]. Moscow, Dialektika Publ., 2007, 912 p.
Dingyu Xue, Yangquan Chen. Solving Applied Mathematical Problems with MATLAB. London, Taylor & Francis Group, 2009, 418 p.
Powers D.M.W. Evaluation: from Precision, Recall and F-measure to ROC, Informedness, Markedness and Correlation. Journal of Machine Learning Technologies, 2011, vol. 2, iss. 1, pp. 37–63.
