As remote work and online learning grow, assessing an individual’s concentration and tension levels has become more challenging compared to face-to-face environments. This study develops a system to classify high-tension and relaxation states using muscle deformation sensor array. Specifically, we collected muscle deformation data from the calf and constructed a classification model using a Support Vector Machine (SVM). Data were collected from 20 participants (10 males and 10 females) performing five tasks: relaxation, mental arithmetic, writing, watching, and conversation. Initially, a five-class classification (relaxation, mental arithmetic, writing, watching, conversation) was performed. However, by consolidating tasks with similar muscle deformation patterns into a three-class classification (relaxation, high-tension, conversation), the classification accuracy improved. Furthermore, in the binary classification of high-tension and relaxation states, individual models were created for each participant, achieving a high classification accuracy of 99.1%. Using individual models allows for more precise estimation of tension levels by considering variations in muscle deformation patterns among participants. This study demonstrates the potential for real-time classification of high-tension and relaxation states using optical muscle deformation sensors, with applications in productivity enhancement and stress management in remote work and educational environments.