Artificial Flora Optimization Algorithm with Genetically Guided Operators for Feature Selection and Neural Network Training

Many real-life optimization problems in different fields of engineering, science, business, and economics are challenging to solve, due to the complexity and as such they are classified as non-deterministic polynomial-time hard. In recent years, nature-inspired metaheuristics are proved to be robust solvers for global optimization problems. Hybridization is a commonly used technique that can further improve metaheuristic algorithms. Hybrid algorithms are designed by combining the advantages of various algorithms, which produce a synergistic effect. Hybridization results in intensifying specific advantages in different algorithms and the hybridized algorithm implementation often performs better than the original version. In this paper, we present the hybridized artificial flora optimization algorithm, named genetically guided best artificial flora. This hybridization is achieved by using a uniform crossover and mutation operators from the genetic algorithms that facilitate exploration of the search space and make the right balance between diversification and intensification. Furthermore, the proposed hybrid algorithm is adopted for two real-world problems, artificial neural network training, and feature selection problem. Following good practice, proposed method was first tested on standard unconstrained functions before it was evaluated for these two very important machine learning challenges. The experimental results show that the proposed hybridized algorithm is highly competitive and that it establishes a better balance between exploration and exploitation than the original one and that it is superior over other state-of-the-art methods in artificial neural network training and feature selection.

Automated summary

This study proposes a hybridized artificial flora optimization algorithm named genetically guided best artificial flora, achieved by combining genetic algorithms' uniform crossover and mutation operators. The hybrid algorithm shows excellent performance in artificial neural network training and feature selection, striking a better balance between exploration and exploitation compared to the original version.