Pulse diagnosis is one of the diagnostic methods in traditional Chinese medicine (TCM). Such diagnoses are made subjectively by a TCM doctor, who requires expert knowledge. If pulse diagnosis could be automated, it would be beneficial for health management. In our previous study, we showed that pulse diagnosis might be related to personal health data, such as step count and sleep score. In this study, we propose a new approach to classifying pulse diagnoses based on a combination of features from pulse and health data. Pulse characteristics are extracted from electronically recorded pulse shapes, and health data feature analysis is augmented by considering the periodicity of daily health metrics. Using these features, we perform both single- and multi-label classifications, and investigate the possibility to improve classification accuracy. We further adopt two classification methods for multi-label classification: random forests and deep learning. Our results show that our approach improves classification accuracy for pulse diagnoses.