aManipal University Jaipur, Jaipur 303007, Rajasthan, India
bSir Padampat Singhania University, Udaipur 313601, Rajasthan, India
cTechno NJR Institute of Technology, Udaipur 313003, Rajasthan, India
Early and precise diagnosis of neurodegenerative (Alzheimer's) disease is essential for efficient intervention and disease management. To improve disease prediction using the OASIS dataset, this research investigates a methodical framework that combines several machine learning classifiers with sophisticated feature selection techniques. We compare three most used feature selection techniques for this research: Correlation-Based Feature Selection (CFS), Wrapper Forward Selection (WFS), and LASSO regression with five classifiers: C5.0, CHAID, Logistic Regression, K-Nearest Neighbors (KNN), and Linear Support Vector Machine (LSVM). We also examine how the Synthetic Minority Oversampling Technique (SMOTE) overcomes the challenge of class imbalance. According to experimental results, LASSO and SMOTE combined with LSVM and CHAID produce better predictive performance, with accuracies of up to 94.66% and 94.34%, respectively. Interestingly, C5.0 achieves a 96.10% peak accuracy.
Index Terms: Dementia, Machine Learning, Feature Importance, Neurodegenerative disease, Alzheimer's.
How to cite this article: Arya AD, Verma SS, Chakrabarti P, Bishnoi R. Advancing Neurodegenerative Disease Prediction: Innovations in Feature Engineering and Machine Learning. Int J Drug Deliv Technol. 2026;16(3): 290-303. DOI: 10.25258/ijddt.16.3.35
Source of support: Nil.
Conflict of interest: None