An Empirical Development of Hyper-Tuned CNN Using Spotted Hyena Optimizer For Bio-Medical Image Classification

Abstract

This paper attempted to obtain the optimized value by tuning the hyper-parameters of Convolutional Neural Network (CNN) by mimicking the biological and collaborative behaviour of Spotted Hyenas. To resolve the complications associated with tuning the hyper-parameters of CNN such as; learning rate, momentum, number of epochs, batch size, kernel size, kernel type, stride, padding, number of nodes in hidden layers, and activation functions a bio-image classification model has been developed empirically by utilizing the explorative strength along with exploitation in the search region using a recently developed meta-heuristic optimization approach inspired by the life of spotted hyenas such as; encircling, hunting, attacking and searching for prey to solve continuous optimization problems by mathematically modeled technique. The suggested method has been assessed by traditional Artificial Neural Network (ANN) based strategies as well as Deep learning network-based strategies. The efficacy of this hyper-tuned CNN classification model has been proven by various accuracy measures, convergence analysis, and statistical implications on two benchmark datasets from Kaggle data repository. From this study, it can be observed and concluded that the suggested nature inspired method of tuning the hyper-parameters of CNN is an effective and trustworthy algorithm that has the ability to classify the brain MRI images in an optimized manner.