Classification Using Neural Network and Genetic Algorithm

Overview

This research focuses on improving classification performance using a hybrid approach that combines Neural Networks and Genetic Algorithms.

Traditional neural networks trained using backpropagation often face challenges such as slow convergence and getting trapped in local minima. To overcome these limitations, this work applies genetic algorithm-based optimization using gene reconfiguration to evolve both network structure and connection weights.

Key Contributions

• Hybrid ANN + GA model for optimized classification
• Evolution of neural network architecture and weights
• Faster convergence and improved training efficiency
• Reduced risk of local minima during training

Methodology

• Genetic Algorithm used for:
  • Network structure evolution
  • Connection weight optimization
• Neural Network used for classification learning
• Cross-validation applied for robust performance evaluation

Results

The proposed model demonstrated strong performance across multiple benchmark datasets:

• Iris Dataset

  • Training Accuracy: 99.3%
  • Testing Accuracy: 97.3%

• Wine Dataset

  • Training Accuracy: 99.9%
  • Testing Accuracy: 92.9%

• Breast Cancer Dataset

  • Training Accuracy: 98.1%
  • Testing Accuracy: 97.0%

• Diabetes Dataset

  • Training Accuracy: 78.9%
  • Testing Accuracy: 75.5%

• Heart Dataset

  • Training Accuracy: 64.2%
  • Testing Accuracy: 57.6%

Compared to traditional neural network approaches, the hybrid model achieved improved accuracy, better generalization, and more stable performance across datasets.

Conclusion

This research demonstrates that integrating Genetic Algorithms with Neural Networks significantly enhances classification performance. The approach is effective for building robust, scalable, and high-accuracy models in data mining and intelligent systems.