Genetic algorithm/extreme learning machine paradigm for cancer detection

Abstract

Biological systems inspire many machine learning systems. Two of these systems are genetic algorithms (GAs) and neural networks (NNs). A special type of NN is Extreme Learning Machine (ELM). ELM is a single hidden layer feedforward NN, in which the training step is done in just one step. Even if the hidden-output weights are analytically computed, studies have shown that the ELM still maintains the universal approximation capability. In the classical ELM the input-hidden weights are randomly generated. This paper deals with the evolutionary training of an ELM, by using a GA routine to set the input weights. The new hybrid has been applied on two real-world datasets concerning breast cancer detection. The results obtained show that the new technique is competitive to other state-of-the-art methods.