Unstructured data default probability prediction method based on deep learning

Abstract

The invention relates to an unstructured data default probability prediction method based on deep learning. The method comprises the steps as follows: unstructured data, including text data and time series data, of credit subjects are integrated and cleaned; the unstructured data are converted into a data format recognizable by a deep learning model; data features are extracted as sample data on the basis of a deep learning model frame; as for the extracted sample data, a credit risk model is constructed by use of a complex machine learning classification algorithm-integrated tree model, and default probability prediction is output. According to the method, the unstructured data such as text and time sequence data are mined, potential risk behavior modes of the credit subjects are caught on the basis of deep learning and a big data technology, high dimensional data credit risk modeling is performed accordingly, automatic, comprehensive and procedural quantitative credit risk analysis for the credit subjects is realized, the finance risk control capacity is improved, and the credit risk is reduced.

Description

technical field [0001] The invention relates to the field of financial risk control, in particular to a method for predicting default probability based on deep learning and unstructured data. Background technique [0002] With the rapid popularization of the mobile Internet, users' financial behavior habits are undergoing tremendous changes, and most people are becoming more and more accustomed to arranging daily food, clothing, housing and transportation through the Internet in their lives. Affected by this, the volume and richness of unstructured data such as user e-commerce data, behavioral data, and social data have increased by leaps and bounds compared with the past few years. On the one hand, these data have the characteristics of passive data, which are more authentic and difficult to forge, and can objectively describe a person's long-term financial behavior habits; on the other hand, they also have the advantages of easy access and low acquisition costs. Therefore...

AI Technical Summary

Problems solved by technology

[0004] 1. Data cleaning is heavy: Since unstructured data such as text and time-series data are naturally large in size and difficult to process, simple processing methods not only consume a lot of computing resources, but are far from reaching the level that can be processed

[0005] 2. Complex data transformation: In order to use the deep learning model to extract the value of the full amount of data, it is necessary to transform unstructured data into tensor form. Traditional transformation methods have shortcomings such as sparse matrix and excessive loss of information.

[0006] 3. Difficulty in feature extraction: Traditionally, for vectorized data, features are often extracted manually, or directly hard-coded regular expressions on the original data. However, such methods have major limitations, not only the extraction value is limited , and it is difficult to adapt to the rapid evolution of the online risk environment

[0007] 4. Difficulty in risk modeling: Since the features that can be extracted from unstructured data are often thousands or even tens of thousands of dimensions, far beyond the range that traditional scorecard models can handle, financial institutions need more cutting-edge machine learning algorithms to complete Modeling and a range of methods for evaluating and automatically outputting default probability forecasts

Images