MELPD-Detector: Multi-level ensemble learning method based on adaptive data augmentation for Parkinson disease detection via free-KD

Parkinson disease (PD) is a neurodegenerative disorder which has tremor in the finger, handwriting change and so on. Tremor in the finger with PD changes the typing pattern of subjects. Keystroke dynamics-based PD detection is class-imbalanced problem due to the scarcity of PD keystroke data. To address these problems, we propose a novel multi-level ensemble learning (EL) method that incorporates adaptive data augmentation techniques to diagnose PD. Specifically, we propose adaptive data augmentation methods on three base models to solve class-imbalanced problem. Further, we propose multi-level ensemble learning method for different temporal relation between different types of free-text keystroke dynamics (free-KD). Extensive experiments on datasets demonstrate that accuracy of our proposed method is up to 99.8%. In addition, our proposed method has high generalization and robustness on both free composition and transcription tasks.