Abstract: In order to improve the accuracy and generalization ability of dimension model physiological signal emotion recognition, this paper extracts the temporal statistical features, power spectral features and wavelet packet entropy features of GSR signal based on the DEAP dimension emotional physiological data set. Decision tree and random forest algorithm were used to classify emotions on two emotional dimensions of arousal and valence. To improve the accuracy and robustness of machine emotion recognition by selecting appropriate dimension emotion label threshold; Using GSR signal difference between induced emotion state and natural emotion state conduct normalize processing, eliminate individual differences, and improve the generalization ability of the model; Applying multi-features fusion and random forest algorithm based on integrated learning to obtain better emotion recognition performance.The classification accuracy Acc and F1 values were 92.0% and 0.933 in the arousal dimension, and 90.9% and 0.925 in the valence dimension.The simulation results showed that the machine learning method based on tree model can realize the accurate recognition of dimension emotion, which can be used for automatic analysis of physiological signal emotion and machine recognition of wearable devices.