This paper introduces a principal component analysis (PCA) approach to extract principal components from nuclear mass models. The PCA method is used to identify the major patterns in nuclear mass models, which are then recombined to build new mass models that achieve better accuracy than the original theoretical models. The extracted principal components are uncorrelated and ordered by their importance in representing the relevant features of the original models. The first principal component is mainly contributed by bulk properties, as described in the liquid drop model (LDM), while the second is mainly contributed by deformation related to shell effects and odd-even effects. The third and fourth principal components are related to the breaking of neutron and proton symmetry energy. These principal components are then recombined to build new nuclear mass models, which achieve better accuracy in predicting experimental mass data. The results show that the effects contained in different theoretical mass models can work together to improve nuclear mass predictions through PCA. The new mass models built using PCA outperform the original theoretical models in terms of accuracy. The PCA approach is also shown to be effective in avoiding overfitting, as demonstrated by extrapolation validation. The study highlights the potential of PCA as a tool for building more accurate nuclear mass models by extracting and combining the major effects from different theoretical models.This paper introduces a principal component analysis (PCA) approach to extract principal components from nuclear mass models. The PCA method is used to identify the major patterns in nuclear mass models, which are then recombined to build new mass models that achieve better accuracy than the original theoretical models. The extracted principal components are uncorrelated and ordered by their importance in representing the relevant features of the original models. The first principal component is mainly contributed by bulk properties, as described in the liquid drop model (LDM), while the second is mainly contributed by deformation related to shell effects and odd-even effects. The third and fourth principal components are related to the breaking of neutron and proton symmetry energy. These principal components are then recombined to build new nuclear mass models, which achieve better accuracy in predicting experimental mass data. The results show that the effects contained in different theoretical mass models can work together to improve nuclear mass predictions through PCA. The new mass models built using PCA outperform the original theoretical models in terms of accuracy. The PCA approach is also shown to be effective in avoiding overfitting, as demonstrated by extrapolation validation. The study highlights the potential of PCA as a tool for building more accurate nuclear mass models by extracting and combining the major effects from different theoretical models.