Analysis of electroencephalographic (EEG) data has shown to be a useful tool in the fields of Brain-computer interfaces (BCIs), and even business. Recently, deep learning architectures have been implemented, particularly in the interpretation of EEG readings and the knowledge it may impart about mental processes, thanks to the availability of huge EEG data sets and advancements in deep learning. In order to find useful data for brain classification and neuroimaging, Various Analytical Techniques utilized in EEG investigations have turned to machine learning. This research study combines sparse auto-encoders (SAEs) with a post-processing system, which is comprised of a linear system model dependent on the Particle Swarm Optimization (PSO) method as a unique data categorization framework. To boost classification precision, it is to prune datasets of superfluous characteristics. The Q-Learning method is used to built on reinforcement learning to reduce the overall number of features and have obtained the resulting accuracy. Evidence of this strategy leading to better classification performance is shown by experiments on both simulated and actual data. The PSO method provides a metaheuristic strategy for estimating the linear model’s parameters. Using publicly available datasets, the suggested methodology is verified; When contrasted to current best practices, the results are favorable. With some adjustments to the input characteristics, hidden neurons, and output classes, the system may be utilized to address any data categorization problem.