Archive-based coronavirus herd immunity algorithm for optimizing weights in neural networks | JoVE Visualize

Area of Science:

Artificial Intelligence
Machine Learning
Optimization Algorithms

Background:

Supervised learning for feedforward neural networks, particularly multilayer perceptrons (MLP), relies heavily on optimal configuration of weights and biases.
Traditional gradient descent methods for MLP training face challenges like local optima and slow convergence.
Metaheuristic algorithms offer an alternative to overcome gradient descent limitations in complex optimization problems.

Purpose of the Study:

To enhance the Coronavirus herd immunity optimizer (CHIO) using an external archive strategy, creating the ACHIO algorithm.
To apply the enhanced ACHIO algorithm for optimizing the controlling parameters (weights and biases) of MLP neural networks.
To evaluate the effectiveness of ACHIO in improving the classification accuracy of MLPs across various datasets.

Main Methods:

An external archive strategy was integrated into the CHIO algorithm to guide the population toward more promising search regions during evolution.
The enhanced algorithm, termed ACHIO, was employed to train MLP models for classification tasks.
ACHIO's performance was benchmarked against six established swarm intelligence algorithms and the original CHIO using 15 diverse classification datasets.

Main Results:

The proposed ACHIO algorithm demonstrated superior performance, achieving higher classification accuracy than comparative methods on ten out of fifteen datasets.
ACHIO provided highly competitive results on the remaining datasets, indicating its robustness and effectiveness.
The external archive strategy in ACHIO successfully improved the optimization process for MLP parameter tuning.

Conclusions:

The enhanced CHIO algorithm (ACHIO) is a powerful tool for optimizing MLP parameters, leading to significant improvements in classification accuracy.
ACHIO offers a promising alternative to traditional methods for training neural networks, especially in scenarios prone to local optima.
The integration of an external archive strategy is an effective technique for enhancing metaheuristic algorithm performance in machine learning applications.