This paper investigates the use of synthetic data generation to enhance unsupervised learning applied to physiological signals collected from wearable devices. Leveraging raw datasets obtained from Withings smartwatches, the study evaluates the influence of synthetic data on clustering algorithms aimed at identifying behavioral patterns associated with affective disorders, including depression, anxiety, and bipolar disorder. Synthetic datasets are generated using the synthpop library in R and validated through statistical utility metrics. A comparative analysis of K-Means, DBSCAN, and agglomerative hierarchical clustering reveals that synthetic data can augment the original dataset without significantly altering its structure, although a mild homogenization effect is observed. These results underscore the potential of synthetic data to support data-driven mental health research while emphasizing the need for careful validation to preserve clinically relevant variability.