STUDY OF CREDIT RISK ASSESSMENT PERFORMANCE IMPROVEMENT THROUGH XGBOOST ALGORITHM
A good credit risk assessment can maximize the economic benefits derived from credit. The challenge of credit scoring, especially in Indonesia, is to help reduce non-performing loans (NPL). For this reason, a credit scoring algorithm is needed that is able to help identify potential borrowers wit...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Final Project |
| 語言: | Indonesia |
| 在線閱讀: | https://digilib.itb.ac.id/gdl/view/56987 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Institut Teknologi Bandung |
| 語言: | Indonesia |
| 總結: | A good credit risk assessment can maximize the economic benefits derived from credit. The
challenge of credit scoring, especially in Indonesia, is to help reduce non-performing loans
(NPL). For this reason, a credit scoring algorithm is needed that is able to help identify potential
borrowers with high potential to be unable to repay loans. In this final project, an algorithm was
developed to assess credit risk with a data science-based approach. The developed algorithm
applies several methods, namely decision tree, random forest, and XGBoost. The data used for
this project is SIJEKH 1. Classifier creation is carried out in five stages, starts from
understanding business needs, understanding data, preprocessing data, optimizing parameters
and classification, and evaluating. Data preprocessing is done by eliminating incomplete data,
then balancing the data using SMOTE. Classification is done by applying k-fold validation. The
performance of the model between one algorithm and another is compared, the decision tree has
a false positive rate (FPR) of 93%, random forest has FPR of 97%, and XGBoost after
optimization has FPR of 12%.
The XGBoost algorithm performs the best but consumes the most training time. After
optimization, the XGBoost algorithm had a decrease in FPR from 94% to 12%.
|
|---|