Mayfly Algorithm Based Convolutional Neural Network for Human Diseases Recognition System

Convolutional Neural Network (CNN) is a machine learning method which mainly focused on the automatic feature selection and matching of images and has been used for detection and recognition. CNN suffers from hyperparameter selection and overfitting problem and can be solved using an optimization technique. Existing optimization technique such as Mayfly Algorithm (MA) still suffers from initial parameter tuning and had slow convergence behaviour. This research developed a Mayfly Algorithm based on Convolutional Neural Network for pulmonary diseases recognition. The X-ray images which include normal and pulmonary diseases cases were obtained from a repository via www.kaggle.com. The images were pre-processed using cropping, contrast adjustment, histogram equalizer and normalization to obtain good images quality. A Mayfly Algorithm was used to optimize CNN hyperparameters. The developed technique was implemented in MATLAB (R2020a) Software. The results obtained were evaluated using standard metric. The CNN technique average results are 96.0%, 94.6%, 3.7%, 95.4% and 82.4μs while MA-CNN average results are 97.1%, 95.9, 3.0%, 96.7% and 60.7μs for Specificity, sensitivity, false positive rate, Accuracy and Computation time respectively at 0.75 threshold. This shows the effectiveness of optimizing CNN hyperparameters for image recognition.