Research On Combined Model Of Personal Credit Risk Assessment ——Based On Feature Derivation And Oversampling

With the rapid expansion of Internet finance and the great change of personal consumption attitude,the demand for personal credit business is also growing rapidly,and the corresponding credit risk is also increasing.Financial institutions need to deal with and assess the credit risk of lenders in a timely manner.In the existing literature research,there is often less attention to the preprocessing of data before modeling.Most studies only rely on the model itself to improve the classification performance,so it is difficult to make a new breakthrough in the final effect.This prompted me to focus on data preparation.At the same time,for the sake of improvement the effect of the model,this paper constructs a combined model by linear weighting.With the help of XGBoost and Lightgbm models,this paper will focus on the research of whether the data preparation and the construction of the combined model can improve the model classification ability.After the routine data preprocessing is completed,new features are derived through different methods,and samples are over-sampled.In the end,a combined model is established to prove whether these methods can improve the effect.The empirical results show that both feature derivation and oversampling can obviously improve the classification prediction effect of model classifier.If they are combined,the prediction effect of the model will still improve,indicating that the data preparation before modeling can indeed help the prediction from different angles.The construction of the combined model also has obvious improvement effect on the final result,which shows that the means of linear weighted summation is a useful and practical way to improve the effect.