Shallow Learning Versus Deep Learning in Biomedical Applications

Alzheimer’s disease (AD) and Parkinson’s disease (PD) are two distinct types of degenerative neurologic disorders that impact brain function and movement in their own ways. Early detection of PD and AD is crucial for improving the patient’s quality of life. For this reason, in this study, we employed shallow learning (SL) and deep learning (DL) methods to diagnose PD and AD. In the pre-processing stage, electroencephalography (EEG) signals were passed through the Finite Impulse Response (FIR) filter for the diagnosis of both diseases. To further enhance the signals, we then implemented Independent Component Analysis (ICA) on the filter’s output. After this step, the EEG signals were divided into segments. There has been no extraction of characteristics carried out to diagnose AD. The Common Spatial Patterns (CSP) feature extraction was used to acquire EEG properties in the diagnosis of PD. The diagnosis of both PD and AD involved using the same shallow learning algorithms, specifically, AdaBoost (AB), CatBoost, XGBoost, Decision Tree (DT), Gaussian Naïve Bayes (GNB), Linear Discriminant Analysis (LDA), and K-Nearest Neighbors (KNN). As a deep learning algorithm, the diagnosis of PD utilized the Deep Neural Network (DNN), while the diagnosis of AD employed the 2-Dimensional Convolutional Neural Network (2D-CNN). The diagnosing PD was most accurately accomplished using KNN, a shallow learning algorithm, which achieved an impressive accuracy rate of 99.3%. Conversely, when it came to diagnosing AD, the deep learning algorithm known as 2D-CNN outperformed the others, achieving the highest accuracy rate of 94.8%.