Stress is an intensified psychophysiological state of the body that develops in reaction to a demanding or challenging situation. Stress levels can be determined utilizing different biosignals, such as thermal, electrical, impedance, auditory, and optical changes that occur when a person is under stress. When mental health is adequately handled, the quality of human existence can be considerably increased. To prevent a person from numerous stress-related health issues, this research provides various machine learning and deep learning algorithms for stress detection on persons utilizing multimodal datasets gathered through wearable physiological and motion sensors such as photoplethysmography (PPG), galvanic skin response (GSR), and electroencephalography (EEG). The performance for binary (positive, negative) classification and prediction was assessed and compared by using machine learning techniques like random forest, decision tree, support vector machine, and logistic regression. Besides, the recurrent neural network and encoder–decoder network were introduced for binary classifications. During the study, the random forest classifier produced an accuracy of 81% for binary classification problems, F1-score of 0.77, precision of 0.72, and 85% for recall. By using deep learning, the achieved accuracy is up to 95% for anger and joy, 94% for sadness, and 92% for fear and neutral respectively.