The problem of recognizing accurately the pain is a significant challenge in affective computing, mainly because of the individual differences in expressing or hiding pain. Most traditional models fail to consider the internal conflict—cognitive dissonance—between perceived pain and expressed pain. To bridge this inadequacy, we introduce here a new cognitive dissonance-aware pain recognition model capable of incorporating inter-individual differences from multimodal physiological signals. Motion and surface Electromyography (sEMG) features are then extracted and employed to map subjects into a [Pain, Expression] space. We also apply K-Means and Fuzzy C-Means (FCM) clustering to infer underlying subject profiles—Suppressive, Expressive, and Ambiguous—using their expression dynamics. While K-Means provides interpretable and transparent groupings, it imposes hard cluster assignments that could mislabel ambiguous behaviors. FCM, on the other hand, holds soft membership intact and facilitates richer borderline object description and Ambiguous category definition to a wider extent. FCM, however, failed to detect a pure Suppressive cluster, suggesting more fluid boundaries of behavior. Trained classification models of both types of clusters all show improvement in accuracy, F1-score, and Receiver Operating Characteristic (ROC) curve to support the advantage of cluster guided, personalized modeling. The findings emphasize the value of soft clustering in demystifying otherwise mystifying affective states and form the basis for more interpretable and personalized pain recognition systems.