MENTAL STATES AND ALCOHOLISM DETECTION USING NON-STATIONARY DECOMPOSITIONS FOR EEG SIGNALS

Abstract

An electroencephalogram (EEG) is an electrical signal that captures brain impulses by arranging electrodes in a specific pattern. The process of collecting EEG can be either invasive or non-invasive. Non-invasive EEG recordings are obtained from the electrodes attached to the scalp area, whereas invasive EEG recordings are obtained from the electrodes implanted into the brain, which needs surgery. As a result, EEG data provide information into the participant's cognitive conditions. EEG, on the other hand, is susceptible to electrical noise and is best used in controlled lab circumstances rather than in real-world scenarios. Non invasive EEG devices are becoming popular among researchers, and they are most widely used brain signal methods. The non-invasive nature of EEG-based techniques makes it more viable for the following reasons: EEG is a fast and safe way of checking brain activity, EEG approaches are non-invasive, EEG detects brain activity at a resolution of milliseconds with high precision. EEG has a less error probability, convenient to use and less setup cost with minimal danger. The advancements in technology have bought lots of changes in human life. All the changes have reduced the efforts of humans in doing any work. The advancements reduce the mental attention level of human beings which can be dangerous in attention seeking applications. In this work, a brain-computer interface (BCI) system is suggested for mental attention detection using EEG. To detect mental states a signal processing and machine learning-based algorithm is proposed. Flexible Analytic Wavelet Transform (FAWT) explores for feature extraction from EEG and different machine learning algorithms are tested with extracted features to detect mental states. v Initially, multiple FAWT based features are extracted out of which the log energy entropy provides the best classification performance with an optimizable k-nearest neighbor classifier. The classification performance of the proposed work has better results as compared to other similar approaches. Alcohol consumption alters the functionality of nervous system by disturbing the neuron process, which leads to the behavioral changes in a human life. An automatic identification of alcoholics can address these issues. EEG is a widely used tool for monitoring the brain activities. In this study, singular spectrum analysis (SSA) and machine learning-based algorithm is proposed for the automatic detection of normal and alcohol EEG signals. Kruskal Wallis test is performed as a part of a statistical study and the features which satisfy p<0.05 are considered in the classification. Initially, multiple SSA-based features are extracted out of which the inter-quartile range and wavelength provide the best classification performance with an optimizable support vector machine classifier. The achieved classification accuracy is 94.2%. EEG finds various applications in the identification and diagnosis of neurological diseases and brain computer interface. BCI is a system works on the instructions given by the human brain and helps the disabled to communicate with surroundings. Alcoholism is one of the leading causes of disease and can be identified and diagnosed by using the EEG signals which can avoid road accidents. To explore the complexity of the EEG signals the signal processing in machine learning tools are employed. Various non-stationary tools are employed for the decomposition of EEG signals and multiple features are extracted. The extracted features are further tested with various machine learning algorithms for the classification of EEG signal.