Abstract
This project investigates the application of machine learning and deep learning models for classifying gastrointestinal (GI) diseases using the Kvasir dataset, which contains labeled endoscopic images across eight disease categories. Starting with baseline models like Linear Regression and Support Vector Machine (SVM), the study progresses to advanced architectures, including MobileNetV3, VGG19, and InceptionV3, to evaluate their suitability for medical image classification. Each model’s performance is assessed using metrics such as accuracy, loss curves, and confusion matrices, with InceptionV3 demonstrating the strongest generalization and reliability across the GI disease categories.
The findings indicate that traditional models provide a useful starting point but are limited in handling the detailed patterns found in medical images, a task where deep learning models perform well. InceptionV3, followed closely by VGG19, proved to be the most effective for accurate GI disease classification, providing robust and consistent results across classes. This study underscores the effectiveness of advanced deep learning architecture in medical diagnostics, particularly for complex image data.