Fuzzy-GA: Genetically-Optimized TSK Fuzzy Inference System

A modern Java 23 implementation of a Takagi-Sugeno-Kang (TSK) Fuzzy Inference System optimized by a Genetic Algorithm.

1. Project Overview

This project provides a flexible and high-performance framework for solving complex regression and mapping problems using a genetically-optimized fuzzy inference system. It takes a classic neuro-fuzzy system and completely refactors it into a modular, testable, and extensible system using modern Java 21 features.

Core Features

• TSK Fuzzy Model: Implements IF-THEN rules with linear function consequents (z = px + qy + r).
• Einstein Product: Utilizes the Einstein product t-norm for a smooth fuzzy AND operation.
• Genetic Algorithm Optimization: A powerful genetic algorithm tunes all parameters of the fuzzy system, including membership functions and rule consequents.
• Modern Java: Built with Java 21, leveraging features like Records, Sealed Classes, and Streams for clean, performant, and maintainable code.
• Decoupled & Modular: The Fuzzy System, Genetic Algorithm, and the bridge connecting them are in separate, decoupled modules, making the system easy to maintain and extend.
• Configurable & Extensible: The GA is configurable via a GaConfig record, and the system can be extended with new membership functions or GA operators.

2. System Architecture

The system is composed of three primary modules that work in concert: the Fuzzy System, the Genetic Algorithm, and the Core bridge.

graph TD
    subgraph Core
        C(FuzzySystemFitnessEvaluator)
    end

    subgraph GA
        B[GeneticAlgorithmEngine]
    end

    subgraph Fuzzy
        D[TskInferenceSystem]
    end

    A[Dataset] --> C
    B -- Uses --> C
    C -- Decodes Chromosome into --> D
    D -- Produces Output --> C
    C -- Calculates Fitness --> B

    style GA fill:#D5E8D4,stroke:#82B366,stroke-width:2px
    style Fuzzy fill:#DAE8FC,stroke:#6C8EBF,stroke-width:2px
    style Core fill:#F8CECC,stroke:#B85450,stroke-width:2px

Loading

Package Structure

• com.fuzzyga.fuzzy: Contains all components for the TSK fuzzy inference system.
• com.fuzzyga.ga: Contains the GeneticAlgorithmEngine and its components.
• com.fuzzyga.core: The bridge between the fuzzy and GA modules, containing the FuzzySystemFitnessEvaluator.
• com.fuzzyga.app: The main application entry point for running the system.
• com.fuzzyga.utils: Utility classes for loading configuration and data.

3. Conceptual Background

A detailed, non-technical explanation of the core concepts can be found in SYSTEM_CONCEPTS.md.

4. Getting Started

Prerequisites

• Java JDK 23+
• Apache Maven 3.8+

Installation & Build

# Clone the repository
git clone <your-repo-url>
cd <repo-name>

# Build the project, compile sources, and run tests
mvn clean install

Running the Application

The main application class com.fuzzyga.app.App runs the entire optimization process using the sample dataset and configuration.

# Run the main demo application from the project root
mvn exec:java -Dexec.mainClass="com.fuzzyga.app.App"

The application will output the progress of the genetic algorithm for each generation and print the final, best-found solution.

5. Usage & Code Examples

Configuration

The GA hyperparameters can be configured in src/main/resources/config.properties. These parameters control the evolutionary process.

Code Example

The following snippet (simplified from App.java) shows how to wire together the components to run the system.

// 1. Load configuration and data
GaConfig config = ConfigLoader.loadConfig("config.properties");
Dataset dataset = Dataset.fromResource("datasets/3dfunc.txt");

// 2. Describe the desired fuzzy system architecture
InputVariable x = new InputVariable("x");
InputVariable y = new InputVariable("y");
FuzzySystemDescriptor descriptor = new FuzzySystemDescriptor(
    List.of(x, y), // Input variables
    3,             // Number of fuzzy sets per variable
    Map.of(        // Parameter ranges for optimization
        "center", new FuzzySystemDescriptor.Range(-5.0, 15.0),
        "width", new FuzzySystemDescriptor.Range(1.0, 5.0),
        "consequent", new FuzzySystemDescriptor.Range(-10.0, 10.0)
    )
);

// 3. Create the evaluator and the GA engine
FuzzySystemFitnessEvaluator evaluator = new FuzzySystemFitnessEvaluator(descriptor, dataset);
GeneticAlgorithmEngine ga = new GeneticAlgorithmEngine(config, evaluator);

// 4. Run the evolution
Individual bestSolution = ga.evolve(null, null);

// 5. Use the results
System.out.println("Best solution fitness: " + bestSolution.fitness());

6. Future Improvements

• Support for other membership functions (Gaussian, Trapezoidal).
• A wider variety of GA operators (e.g., different selection or crossover strategies).
• A GUI for visualizing the fuzzy sets, rule surface, and GA convergence.
• Support for classification problems in addition to regression.