Application of OLAP based Business Intelligence and Predictive Analysis (Random Forest Classifier and ARIMA) in a Fintech P2P Lending Company

Abstract

The rising rate of non-performing loans (NPLs) in the financial technology (fintech) P2P lending industry has become a serious challenge, including for PT XYZ, which focuses on funding the MSME sector. This study applies Business Intelligence using OLAP and predictive analytics at PT XYZ over a two-year period. The dataset consists of records from January 2023 to December 2024, comprising 276,320 loan transactions and 267,542 repayment records. The OLAP approach was employed to explore loan transactions, repayment performance, and delinquency patterns across multiple dimensions, such as account, time, payment method, and bank used. The results of this exploration were visualized through three interactive dashboards tailored to the analysis focus. Predictive analysis was conducted using a Random Forest Classifier, which identified loan_amount as the most influential variable in predicting potential delinquency. However, the model’s performance was relatively limited, achieving an accuracy of 61.47%, precision of 38.33%, and recall of 36.89%. Delinquency trend forecasting using the ARIMA (2,1,1) model indicated an increase in overdue cases up to March 2025, followed by a decline in subsequent months. The model evaluation yielded MAPE of 39.74%, sMAPE of 33.32%, and RMSE of 1086.77, suggesting moderately adequate predictive accuracy but with a relatively high error rate. This study recommends implementing incentives for OBA (Online Borrower Accounts) with good repayment performance and restricting loan ceilings based on an automated risk score. Further development is advised by enriching the dataset with additional variables and enhancing predictive models through techniques such as oversampling (e.g., SMOTE) and hyperparameter tuning to maximize accuracy.