Stress, which can occur due to several factors, such as work pressure, life changes, worries, over-responsibilities, and experiences of abuse or hate, is a natural human response to changes or challenges, resulting in various physical, behavioral, and emotional reactions. The early identification and management of mental stress are crucial as they can prevent numerous health problems like anxiety, cardiovascular diseases, depression, and other stress-related disorders. Classically, by analyzing physiological signals that are indicative of the body’s stress response, human stress is detected. By examining heart rate variability, heart activity, wave intervals, and other physiological features using Electrocardiogram (ECG) and wearable sensors, conventional techniques detect stress to measure the electrical changes responsive to stress. For stress detection, this review describes the process, merits, and demerits of existing Machine Learning (ML)-centric models. The traditional models aim to provide accurate and real-time assessments of stress levels by using physiological data. These system’s reported detection accuracies vary across studies, generally ranging from moderate to high levels, reflecting the current development and improvement of these models.