Abstract: This study develops an indoor temperature prediction model for residential buildings by combining an infrared camera mounted on an unmanned aerial vehicle (UAV) with a random forest regression model. Experimental and simulated analyses were used to evaluate the effects of the measurement distance, pitch angle, and apartment location on prediction accuracy. Eight households with different locations for UAV-based infrared thermography experiments were selected to obtain valid thermal images and environmental parameters. The UAV measurement results show that the distance does not significantly affect the prediction accuracy. The pitch angle plays a critical role: 0° was the optimal angle for accurate infrared signal reception. The middle-floor central units demonstrated stable temperature distributions and achieved the highest prediction accuracy. In contrast, top floor and corner units, affected by convective heat transfer with outdoor air, exhibited greater temperature fluctuations and proved more challenging to predict, with nonheated units showing the worst prediction performance. The random forest regression model, which integrates thermal image features with environmental parameters, achieved high predictive accuracy and demonstrated excellent generalization capabilities.