Dual Attention-Based Dense Capsule Bidirectional Gated Recurrent Unit for Classifying Tremor Types in Parkinson’s Disease-Affected Patients

Abstract

Listen

Translate article

Parkinson’s disease is a genetic disorder which affects the nervous system, including the body’s nerve-controlled organs. Improvements are still required because it is difficult to reduce tremors without impairing voluntary movement. So, in this study, a novel proposed method is used to classify the types of tremors. First, pre-processing is used to remove undesirable noise from the input signals using digital band pass filter (DBPF), Savitzky Golay digital FIR filter (SGDFF), and adaptive wavelet transform (AWT) approaches. Then, the pre-processed signals are fed into the dual attention-based dense capsule bidirectional gated recurrent unit (DA_DCBiGRU) model. A dense capsule is used to extract features, dual attention is utilized to reduce dimensionality, and BiGRU is used to identify tremor kinds such as resting, postural, and action. Finally, the suggested classifier’s efficiency is increased by fine-tuning the parameters with the improved fire hawk optimizer (IFHO). When compared to other existing approaches, the Python tool’s performance measurements indicate the greatest accuracy rate of 98.2%.