GGA-MLP: A Greedy Genetic Algorithm to Optimize Weights and Biases in Multilayer Perceptron

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Optimization

Background:

• Designing Artificial Neural Networks (ANNs) involves optimizing numerous parameters, notably weights and biases, to enhance classification accuracy.
• Traditional gradient-based optimization algorithms often face challenges with local minima.
• Metaheuristic algorithms are increasingly explored as alternatives to overcome limitations of conventional techniques.

Purpose of the Study:

• To propose a novel learning algorithm, Greedy Genetic Algorithm-Multilayer Perceptron (GGA-MLP), for optimizing weights and biases in multilayer perceptrons (MLPs).
• To enhance the performance of the traditional Genetic Algorithm (GA) through a greedy approach in population generation, crossover, and mutation.
• To evaluate the effectiveness of GGA-MLP in classifying complex, nonlinear input patterns.

Main Methods:

• Implementation of a greedy genetic algorithm integrated with a multilayer perceptron (MLP).
• Application of a greedy strategy for initial population generation, crossover, and mutation operations within the genetic algorithm.
• Experimental evaluation on diverse datasets from the University of California, Irvine (UCI) repository to assess classification accuracy.

Main Results:

• The GGA-MLP approach demonstrated improved performance in classifying nonlinear input patterns.
• Experimental results indicated that GGA-MLP achieved classification accuracy comparable to or better than existing state-of-the-art techniques.
• The greedy enhancements to the genetic algorithm effectively optimized weights and biases for MLPs.

Conclusions:

• The GGA-MLP algorithm offers a robust and effective method for optimizing ANN parameters, particularly weights and biases.
• The proposed approach provides a competitive alternative to conventional optimization techniques for improving ANN classification accuracy.
• GGA-MLP shows significant potential for applications requiring high-performance neural network classification.