STATISTICAL FEATURES ANALYSIS OF EEG DATA FOR ALS DIAGNOSIS USING MACHINE LEARNING ALGORITHMS

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a progressive neurological disease resulting from the dysfunction of motor neurons, making early diagnosis difficult due to the slow progression of clinical symptoms. Early diagnosis is crucial to improving the patient's lifespan and quality of life. Electroencephalography (EEG) signals can be recorded non-invasively and at low cost, making them promising biomarkers for ALS diagnosis. Despite the promising results reported in previous EEG-based ALS diagnosis studies, many existing approaches suffer from limited generalizability and insufficient methodological clarity. This study presents an EEG-based solution for ALS diagnosis using machine learning and statistical feature extraction based on time-domain descriptors, namely mean, variance, and higher-order moments, which reflect ALS-related neural alterations. The dataset contained 63,648 channel-level EEG samples obtained from multi-channel recordings of ALS patients and healthy controls. Using the hold-out validation method, the dataset was divided into 80% training and 20% testing sets. The performance of Ensemble Learning (EL), Support Vector Machines (SVM), K-Nearest Neighbor (KNN), and Decision Trees (DT) machine learning methods was compared. The results showed that the EL algorithm achieved the highest performance. The EL algorithm achieved an accuracy of 94.46% and a specificity of 97.13%, demonstrating strong diagnostic potential for ALS classification.