Levy Flight Pelican Optimization Algorithm based feature selection for Multi-Class Gastrointestinal Disease

Endoscopic image analysis plays a crucial role in diagnosis Gastrointestinal Diseases (GD) by allowing visualization of the inner tissues of gastrointestinal tract. However, quality of gastrointestinal images is often suboptimal and GD classification are complex and require multiple parameters to training, affecting their accuracy. In this research, proposed Levy Flight Pelican Optimization Algorithm (LFPOA) and Ensemble Machine Learning (ML) method for classification using methods like Multi-Support Vector Machine (MSVM), K-Nearest Neighbour (KNN), Random Forest (RF), Decision Tree (DT) and Naïve Bayes (NB). The LFPOA reduces dimensionality, balances exploration and exploitation, leads to a global optimum, helping to select relevant features. The EL combines five machine learning techniques to handle multiple classifiers and produce a single output, often leading to higher accuracy. Initially, data are obtained from Kvasir V1 dataset and pre-processing is performed using Adaptive Histogram Equalization (AHE) for image quality enhancement. Feature extraction is conducted using MobileNetV2 and InceptionV3, which use separable convolution and layer resizing to efficiently learn from GD images. The performance of proposed method achieved a high accuracy of 99.25% on Kvasir V1 dataset, compared to existing methods such as Stacked Long Short-Term Memory (SLSTM) and Mask Recurrent-Convolutional Neural Network (R-CNN).