As metropolitan areas evolve into 24-hour hubs, ensuring nighttime thermal comfort has become critical for enhancing urban livability and public space usability. This study explores the use of machine learning (ML) models and social media data to predict thermal comfort in nighttime metropolitan environments. Using a 5-point thermal sensation vote (TSV) scale, over 70,000 social media posts from five diverse U.S. metropolises were analyzed. Use the LLaVA large language model for data cleaning, incorporating meteorological data retrieved via the OpenWeather API. Multiple classification and regression ML algorithms were tested for tasks. The Random Forest models demonstrated the highest performance. Social media user data can improve the accuracy of ML model predictions to a certain extent. The study also shows the ranking of various features’ importance in the thermal comfort ML prediction model. The findings underscore the importance of integrating demographic and environmental data to enhance prediction accuracy and discuss the role of urban greenery in mitigating urban heat island effects. The research can provide actionable insights for urban planners, architects, and policymakers in designing thermally inclusive public spaces, promoting sustainability, and enhancing nighttime urban experiences.