Building convolutional neural network parameters using genetic algorithm for the croup cough classification problem

Abstract

Croup cough is an infection in the upper airway typically occurs in children from age six month to 3 years. Symptoms of croup cough begin with a normal cold, fever and loud barking makes the child difficult to breath. These symptoms are relatively similar with a recent pandemic SARS-COV2. So, the common symptoms of croup cough and SARS-COV2 is urges the physicians to diagnose the infection at early stage. Typically, clinical professions Computer Aided Diagnose system (CADS) for detecting the abnormalities from chest X-Ray (PA View) and CT images of infants. Most of CADS adopted the deep learning technique for classification of radiograph images due to the its ability in term of accuracy rate. Classification accuracy of deep learning techniques like Convolution Neural Network (CNN) highly relays on the weights of convolution filters and fully connected layer. In this work, we propose the optimized CNN using Genetic algorithm (GA) for classification of croup cough images. This work includes optimizing weights of CNN with different batch size and iterations using genetic algorithm to identify the best weights for the classifier to generate maximum accuracy. The experiments were carried out with croup cough image dataset, and we show the promising performance of proposed method of 88.32% accuracy rate with smaller amount of dataset.