Modal frequency is an important indicator for reflecting the health status of the monitored structure. It has been shown in previous studies that the pattern of this indicator is substantially affected by the fluctuation of the ambient conditions. Directly utilizing this indicator without considering the ambient effect will introduce bias to monitoring result. Thus, it is important to recognize the pattern between modal frequency and ambient conditions, such as temperature and humidity. In this paper, a novel machine learning algorithm is introduced to automatic search the optimal model class in modal frequency-ambient condition pattern recognition based on measured data. By introducing a sophisticated hyperparameterization on the weighting parameter vector in the prediction model, the learning algorithm is capable for conducting continuous optimal model searching. The proposed algorithm is then utilized for analyzing dynamic response measurements of a reinforced concrete building.