Research And Implementation Of Federated Transfer Learning Platform For Big Data Credit

Big data credit refers to a credit evaluation and prediction method for enterprises and individuals by analyzing and mining massive credit-related data and applying modeling methods such as machine learning and deep learning.Federated transfer learning technology combines the characteristics of federated learning and transfer learning,enabling joint modeling of different credit data sources without data leaving the domain,which has good application value in the big data credit scenario.However,in the practical engineering application of federated transfer learning,there are still limitations such as general lack of model structure targeting,limited real-time performance,multiple categories and fast updates of credit big data,difficulty in data collection,and complex machine learning platform service deployment processes with poor interactivity.To address these challenges,this paper focuses on the effective application of the federated transfer learning platform in the context of big data credit scoring,with specific research objectives as follows:Building and implementing a credit risk monitoring model based on federated transfer learning.With the need for federated modeling under the premise of data inconsistency,a credit risk monitoring model was constructed based on the federated transfer learning framework and using credit-related batch and stream integrated data which addressed the joint modeling problem of participating parties under large differences in sample space and feature space.Designing and developing a credit data collection tool that supports batch and stream integration.To address the demand for collecting and preprocessing multiple heterogeneous data sources,a data collection tool that supports batch and stream integration was developed,providing functions such as field mapping,full synchronization,incremental synchronization,and data pre-processing.This tool reduces the difficulty of credit data collection and application,enhances data quality and solves the problem of data collection and dataset preparation before federated modeling.Designing and developing a federated transfer learning platform for big data credit scoring.A full-service platform was constructed that integrates data integration,data pre-processing,federated transfer learning modeling,model prediction,and open inference interface.This resolves the current issues of unclear data authorization differentiation,insufficient support for data synchronization and data processing by existing machine learning platforms,and the complex process of building federated computation nodes.The research results of this paper support the construction and operation of the landmark achievement of the national key R&D program"Technologies for Intelligent Credit Evaluation based on Big Data",the"Big Data Credit Intelligent Evaluation and Development Service Platform" and also provide valuable reference for the product and service design of credit agencies in the era of digital economy.