Deep Forest Network for User Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in effectively identifying abnormal users or transactions during online payment processes, particularly due to the scarcity of abnormal samples, which leads to low network accuracy and inefficiencies in risk control scenarios.
Innovation Solution
An object classification method utilizing feature extraction and segmentation to obtain sub-features, which are then input into a pre-trained deep forest network for decision classification, with the option of k-fold cross-validation to alleviate over-fitting, enabling more accurate identification of abnormal users or transactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature extraction and segmentation is performed to obtain sub-features for input into deep forest network, then prediction accuracy for abnormal users and transactions is enhanced, but device complexity and processing time increase
Solution Approach 1:
The patent segments the initial feature vector into multiple types of sub-features (e.g., user features, transaction features, device features) and processes them through separate decision tree forest sets at different levels of the deep forest network. This segmentation allows the system to handle complex data more efficiently by breaking it down into manageable components that can be processed in parallel, thereby improving prediction accuracy while managing processing complexity.
2Reliability
If k-fold cross-validation is used to train base classifier, then over-fitting problem is alleviated, but training time increases
Solution Approach 1:
The patent applies k-fold cross-validation to train the base classifier, which involves dividing the training data into k subsets and training the model k times, each time using a different subset as validation data. This partial repetition of the training process with different data splits ensures better model generalization and reduces over-fitting, although it increases training time compared to a single training run.
3Difficulty of detecting and measuring
If deep forest network is used for classification, then detection capability for sparse abnormal samples is improved, but network complexity increases
Solution Approach 1:
The patent implements a multi-level deep forest network structure where decision tree forest sets are nested across multiple levels. Each level contains multiple decision tree forests that process different aspects of the input features. This nested structure enables the network to progressively refine its classification decisions, improving detection capability for sparse abnormal samples while organizing the complexity in a hierarchical manner that facilitates training and inference.
Data Source
AI summary
The present disclosure describes techniques for object classification using deep forest networks. One example method includes classifying a user object including features associated with the user based on a deep forest network including identifying one or more user static features, one or more user dynamic features, and one or more user association features from the features included in the user object; providing the user static features to first layers, the user dynamic features to second layers, and the user association features to third layers, the first, second, and third layers being different and each providing classification data to the next layer based at least in part on the input data and the provided user features.