Feature selection with Lasso for classification of ischemic strokes based on EEG signals

Abstract

Electroencephalography (EEG) is an electrical signal data that can describe brain activity in which the signal contains important information that can be used to detect several diseases. One of the diseases that can be detected by EEG signals is stroke. The most common type of stroke is the acute ischemic stroke (AIS) due to blockage of blood supply to the brain which can generate the tissue damage in the brain EEG signal recording uses several electrodes where the more electrodes used in the recording, the greater the number of EEG features produced (high dimensional data). This can make it difficult for models of machine learning to have optimal performance on high-dimensional data. In this study, for optimizing the performance of the machine learning model by selecting features with the Least Absolute Shrinkage and Selection Operator (Lasso) method, where this method can select the relevant features by shrinking some coefficients to zero. The type of classification used in this study is random forest with features used for classification are Brain Symmetry Index (BSI), Delta-Alpha Ratio (DAR), Delta-Theta-Alpha-Beta Ratio (DTABR). The results showed that the Lasso method can optimize the performance of learning machines with an accuracy value of 75% with 24 features out of 45 features.

Enhancing EEG-Based Mental Stress State Recognition using an Improved Hybrid Feature Selection Algorithm

“…Therefore, current state-of-art efforts employ FS as an optimization technique to enhance the classification performance by selecting the optimal feature set [2]. Furthermore, FS has been successfully utilized to solve several classification problems from various domains, for instance, data mining [2], pattern recognition [4][5][6], and other domains where the high dimensionality occurred.…”