Comparative Analysis of Deep Learning and Machine Learning Models for Alzheimer’s and Parkinson’s Disease Classification from Medical Images

Worldwide, 33.9 million individuals have Alzheimer’s disease (AD), with prevalence projected to quadruple in 40 years. Adequate AD diagnosis is essential for prompt therapy. This study compares machine learning (ML) and deep learning (DL) strategies for AD classification using neuroimaging data obtained from the Kaggle Open Access Collection of Imaging Studies dataset (N=7839). This work tests ML and DL algorithms for AD classification using neuroimaging data. Preprocessing MRI data, data augmentation, model training, and performance assessment utilizing accuracy, precision, recall, and F1-score are the steps. ML algorithms like Random Forest, Logistic Regression, SVM, and MLP are compared. Deep learning architectures like EfficientNetB3, EfficientNetV2S, MobileNetV2, NASNetMobile, and ResNet50 are also evaluated. Random Forest had the greatest accuracy (92.58%) and F1-score (0.824) of the ML models tested. ResidentsNet50 has the greatest accuracy of 99.75% on the original dataset, whereas EfficientNetB3 has the best validation accuracy of 95.16%. The work emphasizes method and model selection based on application needs such as precision, computational efficiency, and interpretability vs. accuracy for effective AD diagnosis using ML and DL approaches.