Development of problem-oriented innovative explainable and hybridartificial intelligence models

In this thesis, four problem-focused studies have been conducted. The first study proposes a classification model using deep learning methods to diagnose tomato leaf diseases. It utilizes six different Convolutional Neural Network architectures—Alexnet, Efficientb0, Googlenet, Shufflenet, Resnet50, and Inceptionv3—to classify diseases on tomato leaves. A hybrid deep learning model is created by combining the feature maps from these models and optimized using Neighborhood Component Analysis, achieving an accuracy rate of 99.50%. The second study proposes a hybrid deep learning model for the early diagnosis and classification of brain tumors. This model uses a Convolution Neural Network-based approach to automatically classify three types of brain tumors: Glioma, Meningioma, and Pituitary. Features from two pre-trained models are combined, and the most effective features are selected using the Relief method before classification is performed with a Support Vector Machine. Experiments show that the model achieves an accuracy rate of 93.2% and provides effective results compared to other studies in the literature. The third study aims to enhance explainability in the diagnosis of Parkinson's disease using optimization-based artificial intelligence algorithms. In the last study, some optimisation-based classification algorithms for breast cancer presence detection are aimed at explainability and transparency.