An Improved Animal Migration Optimization Algorithm to Train the Feed-Forward Artificial Neural Networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Optimization

Background:

• Training artificial neural networks (ANNs) involves complex weight optimization.
• Gradient-based methods often get trapped in local optima, limiting performance.
• Meta-heuristic approaches offer superior exploration capabilities for ANN training.

Purpose of the Study:

• To develop an improved animal migration optimization (IAMO) algorithm for training multilayer perceptrons (MLPs).
• To enhance the IAMO algorithm with Lévy flight for better exploration and escaping local optima.
• To evaluate the performance of the proposed IAMO-MLP algorithm on benchmark and real-world classification tasks.

Main Methods:

• Development of the IAMO algorithm, incorporating Lévy flight for improved optimization.
• Training multilayer perceptrons using the proposed IAMO-MLP hybrid algorithm.
• Validation against benchmark functions and comparison with ten other algorithms on classification problems.

Main Results:

• The IAMO algorithm demonstrated successful escape from local optima.
• IAMO-MLP showed robust performance, independent of initial weight positions.
• The enhanced algorithm significantly outperformed the original implementation on benchmark functions.
• IAMO-MLP achieved competitive results on classification tasks, evaluated by mean squared error and accuracy.

Conclusions:

• The developed IAMO algorithm is effective for training multilayer perceptrons.
• The IAMO-MLP hybrid approach provides a robust solution for ANN training, overcoming limitations of gradient methods.
• The proposed method offers a promising alternative for complex optimization problems in machine learning.