Stress has become prevalent in today’s fast-paced lives, leading to numerous physical and mental health issues. Thus, its detection and intervention at the preliminary stage are crucial to protect a person from its adverse effects. Additionally, labelling of physiological signals collected during an experiment for supervised training is unfeasible owing to the high false-alarm rate. Hence, to mitigate the limitations of supervised methods that rely on labelled datasets and predefined stress thresholds, this study addresses the detection of stress by employing unsupervised machine learning (ML) approaches using heart rate data of patients. It allows for the automatic detection and characterization of stress patterns without prior knowledge or labelled data. Different ML algorithms, including K-mean clustering, spectral clustering, agglomerative clustering, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), were employed to extract meaningful patterns by performing experiments on the stress and well-being in the Early Life Knowledge Work (SWELL-KW) dataset. The analysis and results of this comparative study demonstrate the potential of unsupervised learning for the development of noninvasive, continuous, and robust methods for the detection and monitoring of stress using heart data.