Research On Prepayment Prediction Of Online Lending Based On Machine Learning

Internet-based lending has become a popular funding alternative for organisations and individuals due to its simple procedures,no need collateral requirements,and prompt lending.Prepayment is more common than loan defaults in loan transactions,according to current online lending research,which focuses on loan default characteristics and predicts loan defaults.Prepayment,like loan defaults,will result in a direct loss of revenue for financial institutions,but it will also raise the risk of reinvestment by financial institutions,which is counterproductive to asset allocation.Predicting prepayment behaviour in online lending can help financial institutions adjust their loan risk assessments and improve capital allocation efficiency.In this study,we apply machine learning algorithms to estimate loan early repayment behaviour and then improve the capital allocation capabilities of financial institutions.Its main responsibilities are as follows:The SVM-RFE-XGBoost based loans early repayment prediction model was designed to address the difficulties of large dimensional and imbalanced loan datasets.To begin,in the feature selection phase,the filtered feature selection methodology and the SVM-RFE feature selection method are coupled to pick the best feature subset and eliminate the influence of redundant features on model prediction.Finally,in the classification and prediction stages,the weighted cross-entropy loss function is used as the XGBoost model's loss function,and the weight coefficients of the weighted cross-entropy loss function are used to improve the model's classification ability for a small number of samples,improving the model's overall classification ability.In contrast to the traditional stacking algorithm,which focuses on the limited prediction performance of a single model while ignoring the differences in different base classifiers in unbalanced data sets,a stacking algorithm based on F1 values optimization is proposed to effectively distinguish the prediction ability of base classifiers on unbalanced data sets.The F1 values are the precision and recall evaluation indices.The index F1 values,which take precision and recall into account while categorizing imbalanced data,may represent the model's performance on the unbalanced data set.F1 values are used as a weight to distinguish the classification performance of the basic classifier.The efficiency of this method has been established through comparative trials.The prediction impacts of the two loan prepayment prediction models are confirmed and assessed using the public data set lending club.The experimental results suggest that the model proposed in this study may better anticipate loan prepayment behavior,allowing financial institutions to improve their asset allocation capabilities.