Development of a parallel cascaded adaptive ensemble framework with feature engineering process for the prediction of neurodegenerative disease using temporal clinical data

Neurodegenerative disorders, such as Alzheimer’s, dementia, and Parkinson’s disease, pose a significant worldwide health concern owing to their progressive and irreversible nature. Early and precise prediction of these disorders is critical for prompt intervention and better patient outcomes. To overcome this issue, this paper offers the parallel cascaded adaptive ensemble framework (PCAEF) for forecasting neurodegenerative illnesses based on temporal clinical data from MRI scans. The proposed system combines random forest, support vector machine, and XGBoost into a stacking-based ensemble architecture. Additionally, principal component analysis is used for feature selection and dimensionality reduction, allowing the model to preserve the most important components while decreasing noise and redundancy. Experiments conducted on cross-sectional and longitudinal MRI datasets demonstrate that the PCAEF framework achieves an accuracy of 91.28% for cross-sectional data and 91.05% for longitudinal data, significantly outperforming deep learning baselines such as long short-term memory (LSTM) and convolutional bidirectional LSTM (ConvBiLSTM), which achieved approximately 60% accuracy under the same experimental settings. These results highlight the effectiveness of combining feature engineering with adaptive ensemble learning for clinical prediction tasks involving limited and heterogeneous datasets. Furthermore, the PCAEF framework shows strong potential for future extension to multimodal neurodegenerative disease detection, incorporating additional biomarkers such as EEG, PET, and clinical variables. Overall, PCAEF provides a scalable, robust, and interpretable diagnostic framework for early prediction of neurodegenerative diseases.