The thermal comfort plays very important role in the classroom which directly affect the productivity and academic performance of the students. The traditional thermal comfort model failed to capture the diversity of human characteristics in the dynamic environments like classrooms. This study used ASHRAE Global Thermal Comfort Database II to understand whether the occupant feedback, environmental measurements or combination of both offers the most reliable guide for flexible HVAC control. The study focused on six key records: Thermal preference vote, thermal sensation vote, thermal comfort vote, thermal acceptability vote,humidity and temperature. Those records are standardized and reduced via Principal-Component Analysis (PCA). Using K-medoids algorithm generated 3 clusters. A Cool-acceptable group, a Neutral-warm group, and a Warm-discomfort group. Supervised learning models were trained using Random-Forest algorithm. Models trained on survey question which achieved the accuracy of 83% and the model based on environmental variable, which achieved 71% accuracy. Thermal preference is identified as the key predictor for thermal comfort using Feature importance. The finding shows that, instead of considering the environmental factors alone for thermal comfort combining both thermal preference and basic environmental factors provide a robustness for the comfort prediction. It helps to develop an energy-efficient classroom climate control system.