To optimize the heating control system in a smart home, it is necessary to have a tool that allows you to determine the optimal air heating time. This study is dedicated to the synthesis of the model of the regression dependence of air heating time on the parameters of the heating system and the internal and external parameters of the room. The research justified and derived mathematical expressions for structural and parametric identification of models based on the linear method of least squares based on machine learning. The expediency of using ensembles of models based on decision trees and on the basis of bagging and boosting is substantiated. It is noted that these models have high predictive power and have proven themselves well in the case of small samples. Three types of prognostic models were built and analyzed. For the three investigated heating devices, a trio of the above models was built and trained. The results show that the nature of the heating process is similar in all cases, but the degree of influence of external weather conditions is different. Conditions and restrictions for using models are defined.