Gastrointestinal Disease Classification using Mayfly Optimization Algorithm based Deep Belief Network

Recently, the detection and classification of gastrointestinal disease tract and disorders based on utilizing endoscopic image classification have increased from the biomedical sector. However, the existing classification methods utilized for gastrointestinal classification has limitations where the models face difficulties in distinguishing between certain classes that results in misclassification. To overcome this limitation, a Mayfly Optimization Algorithm based Deep Belief Network (MOA-DBN) for hyper parameter tuning to classify the gastrointestinal disorder accurately. The proposed MOADBN classifies the gastrointestinal diseases effectively with the help of categorical cross entropy for multi-class classification. The mayfly optimization algorithm addresses the limitations of DBN such as sensitive to initialization and complex hyper parameter tuning that enhanced the classification performance. The endoscopy images are acquired and preprocessed to remove noise to improve classification performance. Features with most appropriate information about the gastrointestinal disease are extracted using pre-trained Visual Geometry Group (VGG) model and classified by the proposed MOA-DBN method. Experimental results of the proposed MOA-DBN method achieved accuracy of ${9 8. 8 3 \%}$ which is higher than the existing classification approaches namely Generative Adversarial Network (GAN) and Spatial attention Convmixer.