Development of a Chaotic Tent Map-Based Pelican Optimization Algorithm for Hyperparameter Optimization

Abstract

Metaheuristic optimisation algorithms have received a lot of attention due to their ability to solve complicated optimisation problems without using any gradient information. However, the performance of these algorithms may be affected by insufficient exploration, premature convergence and getting stuck in local optima. The Pelican Optimisation Algorithm (POA) is a novel nature-inspired metaheuristic algorithm with promising optimisation capabilities; this study uses a Chaotic Tent Map-Based Pelican Optimisation Algorithm (CTM-POA) to enhance the exploration and exploitation capabilities of the traditional POA. The paper also explores the convergence behaviour of CTM-POA and compares its optimisation performance to that of the regular POA. The proposed approach is evaluated on a collection of benchmark functions with diverse features, including uni-modal and multi-modal optimisation tasks. The performance is evaluated on the basis of best fitness value, mean fitness value, standard deviation, worst fitness value, convergence rate and computing time. The findings show that the CTM-POA exhibits higher search ability, increased convergence speed or stability, and better optimisation accuracy than the standard POA on selected benchmark issues. The results also show that CTM-POA was more accurate and more stable with lower average best fitness of 0.04, average mean fitness of 1.61 and average standard deviation of 1.62. The CTM-POA also provides a more efficient optimisation framework that can be employed afterwards for hyperparameter optimisation of machine learning algorithms.