Research And Implementation Of Intelligent Collection Platform Based On LSTM

Rapidly developing in recent years,China’s Internet finance industry has seen financial borrowing and lending products such as "Huabei","Jiebei",and "Baitiao" become deeply ingrained in the lives of its citizens.Together with traditional bank loans and other businesses,it has formed a series of huge financial lending industries.However,due to the lack of business model capacity,insufficient repayment ability or willingness of consumers,coupled with the deterioration of the economic environment from the epidemic in 2020 and other multiple factors,the expected amount of debt data in the industry has increased exponentially year by year.Only relying on traditional manual collection is not enough to meet the huge amount of collection business in the market.Under the background of artificial intelligence and big data,intelligent collection platform is gradually used by financial enterprises.Its core function is how to accurately and efficiently determine the risk level of the customer to be collected.Different risk levels correspond to different collection modes.Based on rich background information such as customer behavior,text,image and voice,customers’ risks can be accurately predicted,so as to improve the recovery rate of non-performing assets of financial enterprises.This paper studies the LSTM repayment probability prediction model and method based on converged Attention mechanism,and on this basis,develops and implements an intelligent collection system based on LSTM model,which mainly includes the following research contents:1.The data preprocessing and machine learning modeling methods for repayment probability prediction are studied.Firstly,the model data set is constructed and exploratory data analysis is carried out,and the data is preprocessed,including missing value filling,outlier removal and feature mapping,which lays a good foundation for the construction of the prediction model.This paper studies the construction method of user repayment probability prediction model based on machine learning modeling,including Light GBM model and Prophet model.The experiment shows that the ACC value of Prophet model is 0.99 percentage points higher than Light GBM model in repayment probability prediction,and the Recall value of Light GBM is 1.31 percentage points higher than that of Prophet in repayment probability prediction.Tree model(Light GBM)can better fit objective common sense,while Prophet model makes up for the lack of attention to temporal characteristics of tree model.2.AT-LSTM repayment probability model based on attention mechanism is proposed.Based on the analysis of the LSTM model,the Attention mechanism is introduced to act the Attention memory unit in the middle of the two LSTM layers.The distribution state of the output LSTM of the upper Layer is collected and transmitted to the input of the lower LSTM to achieve the effect of parameter allocation of the input features of the lower LSTM.The experiment shows that in terms of repayment probability prediction,the Recall value of the AT-LSTM model is 3.36 percentage points higher than Light GBM,4.67 percentage points higher than Prophet,1.15 percentage points higher than LSTM,and the ACC value is 2.09 percentage points higher than Light GBM,1.1 percentage points higher than Prophet,and 0.99 percentage points higher than LSTM,which proved that the accuracy of the AT-LSTM model,the traditional machine learning model and the basic LSTM model in repayment probability prediction was significantly improved.To better support repayment forecasting and collection of customers.3.An intelligent collection system based on LSTM model is designed and implemented.According to the AT-LSTM model proposed in this paper,the intelligent collection system is developed based on Spring Boot development framework with Java as the development language and Python processing model side as the data basis.The proposed long-term and short-term memory neural network’s practical value in credit collection problems is demonstrated by these results,and the implementation of an intelligent collection system based on LSTM confirms the model’s feasibility.It provides a model reference for financial institutions in predicting the probability of user repayment and has an important significance for the development of intelligent collection platform.