Fraud Detection Model With Dynamic Layers and Feature Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current financial fraud detection processes are susceptible to manipulation and provide inaccurate results due to reliance on predefined attributes and handling of imbalanced datasets with missing values.
Innovation Solution
A system that automatically selects optimal attributes using optimization algorithms to extract features for financial data classification, dynamically determines the number of layers in a fraud detection model, and classifies data using a residual neural network to identify fraudulent financial statements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If predefined attributes and rules are used for fraud detection, then the detection process is simple to implement, but the detection accuracy is low and susceptible to manipulation
Solution Approach 1:
The patent replaces the mechanical rule-based system with a neural network-based intelligent system. The fraud detection model uses machine learning algorithms to automatically learn patterns from financial data, substituting the manual predefined rules with an adaptive computational system that can detect complex fraud patterns without being explicitly programmed with specific detection rules.
Solution Approach 2:
The patent dynamically adjusts model parameters including the number of layers, learning rate, batch size, and dropout rate during training. This allows the system to optimize its detection accuracy by changing parameters based on the specific characteristics of the financial data being analyzed, rather than using fixed predetermined parameters.
2Measurement precision
If optimization algorithms are used to select attributes dynamically, then detection accuracy is improved, but computational complexity and training time increase
Solution Approach 1:
The patent performs attribute selection and feature extraction during the training phase before actual fraud detection. By pre-processing the data and selecting optimal attributes in advance, the system reduces the computational burden during real-time detection, achieving high accuracy without excessive complexity during operational use.
Solution Approach 2:
The patent extracts only the most relevant features from the financial data using optimization algorithms during training. By selecting and extracting only the critical attributes needed for fraud detection, the system reduces the dimensionality of the problem and decreases computational complexity while maintaining high detection accuracy.
3Productivity
If the fraud detection model is trained with imbalanced datasets, then all available data is utilized, but classification results become inaccurate and unreliable
Solution Approach 1:
The patent applies oversampling to the minority class (fraudulent samples) to balance the dataset. By generating additional synthetic fraudulent samples through techniques like SMOTE (Synthetic Minority Over-sampling Technique), the system ensures that the minority class is adequately represented in training, allowing the model to learn effective fraud patterns without being dominated by the majority class.
4Productivity
If missing values are handled by simple imputation methods, then processing is fast, but detection reliability decreases
Solution Approach 1:
The patent replaces simple statistical imputation methods with a neural network-based imputation approach. The fraud detection model learns to infer missing values from patterns in the available data, substituting basic mean/median imputation with an intelligent system that can predict missing values based on relationships between features, thereby maintaining both speed and reliability.
Data Source
AI summary
A computer system for classifying financial data as fraudulent can include: one or more processors; and non-transitory computer-readable storage media encoding instructions which, when executed by the one or more processors, causes the computer system to: receive a financial data set associated with an organization; automatically select optimal attributes of the financial data set using an optimization algorithm to extract optimal features required to classify the financial data set; dynamically determine a number of layers of a fraud detection model while training the fraud detection model with the financial data set and the optimal features; and classify the financial data set to indicate fraud by executing the fraud detection model in the number of layers using the optimal features.