Tuning Extreme Learning Machine by Hybrid Planet Optimization Algorithm for Diabetes Classification

Abstract

This paper explores hyperparameter optimization and training of extreme learning machines (ELM) applied to diabetes diagnostics. Early detection of diabetes is vital, as timely treatment significantly improves the quality of life of those affected. One of the toughest challenges facing artificial intelligence (AI) is the selection of control parameters suited to the problem being addressed. This work proposes a metaheuristics-based approach for adjusting the number of neurons in a single hidden layer of an artificial neural network in an ELM, as well as the selection of weights and biases (training) of every neuron in the hidden layer. Additionally, an exploration of the planet optimization algorithm’s (POA) potential for addressing NP difficult tasks is conducted. Through the process of hybridization with the firefly algorithm (FA), the POAs’ performance is further improved. The resulting algorithm is tasked with selecting optimal control parameter values for an ELM tackling diabetes diagnostics. A comparative analysis of the ELM tuned by the proposed PAO firefly search (POA-FS) metaheuristics with other state-of-the-art algorithms tasked with the same challenge strongly indicates that the suggested ELM-POA-FS displays superior performance, clearly outperforming contemporary algorithms tackling the same task that it was tested against.