Machine Learning vs. Deep Learning: A Comparative Analysis for Early Detection of Neurological Disease Progression

Multiple sclerosis (MS) is a chronic neurological disease with significant physical and cognitive impacts, making early detection critical for effective management. This study explores various machine learning (ML) and deep learning (DL) algorithms for predicting the conversion from clinically isolated syndrome (CIS) to MS, using clinical and MRI data. Algorithms assessed include Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Decision Trees (DT), Random Forests (RF), Logistic Regression (LR), Gaussian Naive Bayes (Gaussian NB), XGBoost (XG), TabTransformer (2021), and Hypermixer (2022). Performance was evaluated using accuracy, precision, recall, and F1-Score. Among traditional ML models, Random Forest achieved $85 \%$ accuracy. In deep learning, TabTransformer (2021) achieved similar performance, while Hypermixer (2022) improved accuracy to $87.2 \%$. An ensemble stacking approach combining TabnetClassifier RandomForestClassifier, and XGBClassifier with Logistic Regression as the meta-learner achieved the highest accuracy of $98.6 \%$, with precision, recall, and F1-Score all exceeding $99 \%$. The results demonstrate the potential of ensemble methods and DL models for early MS prediction. Future work will focus on refining these models and integrating additional data sources such as genetic markers and advanced imaging to enhance diagnostic precision and clinical applicability.