신용평가모형과 설명가능한 인공지능을 활용한 채권회수 예측모형 비교 연구
A Comparative Study on the Bond Recovery Prediction ModelUsing Credit Scoring and Explainable AI
김민정(KB데이타시스템); 강금석(한국과학기술원)
40권 3호, 1~18쪽
초록
Currently, there is no standardized prediction model for debt collection agencies to predict the probability of debt recovery, resulting in bond evaluations being based on the subjective assessments made by individual debt collectors. The development of debt recovery prediction models offers several advantages, including the establishment and implementation of an allocation strategy, the provision of customer solutions, and the creation of a commission (i.e., a fee paid to the debt collectors) payment system. This study proposes three models for predicting debt recovery that offer varying advantages based on explanatory power and predictive ability. Firstly, it is suggested that Credit Scoring (CS), a scoring model that has been widely used in the financial sector for over 60 years, can also be applied to develop a debt recovery prediction model for debt collection agencies and has excellent explanatory power. Secondly, the latest machine learning algorithms are utilized to enhance predictive power, while eXplainable AI (XAI) techniques are employed to augment the explanatory power of the machine learning model. Thirdly, a new model is proposed that combines the variable transformation technique of the credit scoring model (CS) with machine learning algorithms to improve predictive performance while retaining explanatory power. This new model also demonstrates a meaningful distinction between the recovery amount and recovery rate, which are the primary business goals of debt collection agencies.
Abstract
Currently, there is no standardized prediction model for debt collection agencies to predict the probability of debt recovery, resulting in bond evaluations being based on the subjective assessments made by individual debt collectors. The development of debt recovery prediction models offers several advantages, including the establishment and implementation of an allocation strategy, the provision of customer solutions, and the creation of a commission (i.e., a fee paid to the debt collectors) payment system. This study proposes three models for predicting debt recovery that offer varying advantages based on explanatory power and predictive ability. Firstly, it is suggested that Credit Scoring (CS), a scoring model that has been widely used in the financial sector for over 60 years, can also be applied to develop a debt recovery prediction model for debt collection agencies and has excellent explanatory power. Secondly, the latest machine learning algorithms are utilized to enhance predictive power, while eXplainable AI (XAI) techniques are employed to augment the explanatory power of the machine learning model. Thirdly, a new model is proposed that combines the variable transformation technique of the credit scoring model (CS) with machine learning algorithms to improve predictive performance while retaining explanatory power. This new model also demonstrates a meaningful distinction between the recovery amount and recovery rate, which are the primary business goals of debt collection agencies.
- 발행기관:
- 한국경영과학회
- 분류:
- 경영학