Customer Loan Credit Risk Prediction Based on Improved Sparrow Search Algorithm Optimised Random Forest Algorithm

This paper aims to improve the accuracy of credit risk assessment by optimising the traditional random forest model through the sparrow search algorithm. The study used Pearson correlation analysis to explore the extent of association between customer indicators and loan delinquency. The results show that income, homeownership status, and years of employment in the organisation are positively associated with non-delinquency, while age, loan purpose, amount, interest rate, percentage of income, length of credit history, and approval status are negatively associated with loan delinquency. The model achieved 99.87% accuracy in the training set after 15 iterations. The confusion matrix showed 20,021 customers were successfully predicted with only 26 prediction errors. The test set accuracy was 82.55% with 7092 customers successfully predicted and 1499 prediction errors. In summary, the study achieved significant results in credit risk assessment and provided financial institutions with a more accurate basis for decision-making.