Neural Network-based Data Mining and Fuzzy Logic Expert System
Unlock the wealth accumulated in your data.
Discover hidden patterns and rules that drive your business.
Catch unusual transactions.
NeuroFLExSys General Description
Applications
Neuro-FLExSys
A unique software package for implementing self-learning expert systems. Trained on historical data, it becomes an effective Data Mining tool for discovering patterns and detecting anomalous transactions.
Fuzzy Logic Expert Systems
Ideal for business areas requiring prioritization and resource scheduling. Successfully applied in manufacturing process control and robotics for managing complex, rule-based systems
Neural Networks
Excel at pattern matching, function approximation, and optimization. Their core strengths include advanced classification and clustering tasks for uncovering insights from data.
System description
A Common Foundation
Neural networks and fuzzy logic control systems are both numerical model-free estimators and dynamic systems. They share the common ability to deal with difficulties arising from uncertainty, imprecision, and noise in the natural environment.
The NeuroFLExSys Hybrid
NeuroFLExSys is a neuro-fuzzy hybrid that implements a fuzzy logic expert system as a neural network. This combines the strengths of both: the neural network enables learning and noise tolerance, while the fuzzy logic ensures a manageable, transparent structure.
Advanced Integration and Benefits
In addition, NeuroFLExSys allows for the introduction of high-level thinking into neural networks by manually presetting some expert rules. This combination leverages the strengths of both approaches for more robust and interpretable intelligent systems.
Training
NeuroFLExSys Overview
A multiple-input-multiple-output (MIMO) system trained on historical input/output data. Its training is a structured, multi-step process to build an accurate and efficient expert system.
Training & Rule Formation
The process begins by splitting variable ranges into sets. Unsupervised learning via Kohonen Feature Maps organizes these input/output sets. Rules are then trained through competitive learning, with weak rules being eliminated based on calculated weights.
Optimization & Fine-Tuning
A specialized procedure combines rules to reduce redundancy. Finally, the entire system is refined through supervised learning, using the Back Propagation algorithm to minimize the quadratic error function for optimal performance.
Implementation
Technical Implementation
The system is programmed in .NET language and its parameters are stored in SQL Server database. Database structure allows to analyze derived rules as data clusters.
Data Architecture
All system parameters and models are stored and managed within a Microsoft SQL Server database.
Analytical Capability
The database structure is specifically designed to enable the analysis of derived rules as identifiable data clusters.
Solutions
AR Escalation
Out of Pattern Credit Card Transaction detection
