Mental stress is an issue that creates functional limitations in the workplace. Chronic stress leads to a number of psychophysiological sicknesses. For instance, it raises the risk of depression, heart attack, and stroke. According to the most recent findings in neuroscience, the human brain is the primary focus of mental stress. Perception of biological motion in the human brain determines the risky and stressful situations. Neural signaling of the human brain is used as an objective measure for determining the stress level of a subject. The oscillations of electroencephalography (EEG) signals are utilized for classifying human stress. EEG signals have a higher temporal resolution and are rapidly distorted with unwanted noise, resulting in a variety of artifacts. This study utilizes Extended Independent Component Analysis based approach for artifacts removal. A Multiclass Common Spatial Pattern-based moving window technique is proposed here to obtain the most distinguishable time segment of EEG trials. BiLSTM is used to improve classification results. In order to validate the model performance, two publically available datasets (i.e., DEAP and SEED) are utilized. Employing these datasets, the proposed model achieves state-of-the-art results (93.1, 96.84%) for EEG signal classification to identify stress.