The Application of Machine Learning Algorithms to the Classification of EEG Signals

In order to study cognitive states and detect deficiencies in thinking, planning, and judgement, automated cognitive assessment systems have been developed. Having access to the opportunity to collaborate with experts in their working environment is another benefit of these platforms, particularly in services where decisions have direct effects on people's lives. This dissertation assesses the brain's reaction to mental tasks including motor imagery, attention, working memory, mental fatigue, and attention deficit hyperactivity disorder (ADHD). Electroencephalography (EEG), which encodes neuron activity as action potentials, is a popular method for analyzing these cognitive functions. These EEG data are processed using state-of-the-art machine learning algorithms, automating cognitive ability testing in the process. Recent state-of-the-art approaches have shown problems with high dimensionality, irrelevant channels, subject and task variability, generalizability, and interpretability. Because of this, classifications of intelligence are often inaccurate. The use of meta-heuristics to the solution of both continuous and binary optimization problems was found to be very fruitful in this scenario. Despite increased performance in EEG signal processing for other applications, these methods have seen very little exploration in the field of cognitive evaluation. This encourages more research into these meta-heuristics approaches to intelligence testing. It is important to look for the most appropriate optimization strategy for a certain task, since all of them will not work in every circumstance. In order to address problems in cognitive testing, this research looks on parameter-free optimization algorithms that are tailored to each kind of test. One of the main problems that hinders the effectiveness of ML models is its dimensionality. Another is its lack of generalizability. The overarching goal of this proposed work is to build and construct an effective channel selection and feature selection model making use of meta-heuristic optimization approaches in order to solve these problems in the various cognitive state assessment applications addressed thus far. To properly classify mental reactions to motor imagery, we suggest employing sub-band wavelets and relative PSD features, and we use the harmony search optimization technique to identify the most useful features. The suggested feature selection approach may minimize the dimensionality by as much as 60% by selecting the most important characteristics. Less tuning parameters are required in the suggested strategy. However, tuning becomes obligatory in order to get the desired set of characteristics. In the future implementation of cognitive workload evaluation, this shortcoming is fixed.