Enhanced feature vector reduction of S-transformed electroencephalography signal for optimal Parkinson’s disease detection

Electroencephalography (EEG) has emerged as a valuable tool for detecting Parkinson’s disease (PD) due to its non-invasive nature and simplicity in data collection. By using deep learning and transformation techniques, researchers can extract numerous features from EEG signals, allowing for a detailed analysis of the disease. However, handling such a huge amount of data is computationally demanding, especially for the fabrication of small, portable handheld PD detection devices, which are a dire need due to the alarming rates at which PD is rising. To overcome this challenge, an enhanced version of the artificial butterfly optimization (ABO) algorithm is introduced to select the most significant features from the feature vectors extracted from the time–frequency transformed EEG signals using a deep convolution network. The algorithm improves upon the original ABO by enhancing its ability to explore and refine feature selection. The main improvement is an adaptive mutation rate, which changes dynamically during the process, starting with a higher rate for more exploration in the beginning and lowering it over time for a more focused optimal solution as the training progresses. This allows for a more efficient selection of crucial features, which are then given to a deep neural network to detect PD. Results show that using this modified ABO (M-ABO) paired with a multilayer perceptron as a fitness evaluator leads to faster, more effective performance compared to the original ABO and particle swarm optimization, achieving accuracy rates of 98.24% and 96.85% on two publicly available datasets.