| As the supervision of the consumer finance field becomes more and more mature,small companies suffer a lot.Faced with the high cost and high risk,how to keep quality customers and mine the customer loaning characteristics is particularly important for small enterprises.Based on the data collected by a loan APP,this paper innovatively introduces the PPFS method and a two-step model to predict the loan amount of customers in the next month.As for feature engineering,based on Light GBM model,PPFS method can further mine all kinds of features from the dataset.The result shows that the PPFS method is not only good at analyzing the relationship between the features,but also good at searching for variables that are effective for predicting the loan amount of customers,which finally improve the prediction effect of Light GBM.From the point of view of data distribution,this paper proposes a two-step method to weaken the influence of the unbalance of data distribution.Firstly,it separates the customers into two parts according to whether the customer will borrow money next month.Then the second step will predict the customer's loaning amount of next month based on borrowing customers.The study found that whether compared with a single Light GBM model or an improved model based on the PPFS method,the two-step model based on Cat Boost and random forest has further improved the prediction effect.In addition to the improvement of the model prediction effect,the variables with the highest importance in each model can be obtained according to the importance ranking of the features output by the model.By analyzing the impact of these variables on the amount borrowed by customers,small companies can develop more precise marketing strategies.By classifying customers according to their borrowing behavior,small companies can design more appropriate incentive policies.Both of these methods help small companies allocate resources rationally,reduce costs and increase profits. |
PDF Full Text Request