As an important branch of mathematical statistics, multivariate statistical analysis has developed rapidly, its theory is more rigorous, its content is more solid, and it has a wide range of practical application value. As one of the dimensionality reduction techniques of multivariate statistical analysis, principal component analysis transforms many highly correlated variables into mutually independent or unrelated variables, simplifies the multi-variable high-dimensional space problem into a low-dimensional comprehensive index problem, and reflects the information of the original variables as much as possible through the comprehensive variables. In this paper, a mathematical model of principal component analysis (PCA) is constructed, and some improvement measures are proposed in terms of processing raw data, KMO test and Bartlett spherical test. Laboratory safety evaluation indicators were used to collect original data by expert scoring method, and Matlab software was used to simulate the improved algorithm proposed in this paper. The original 20 evaluation indicators were replaced by 4 main factors, and factors replaced the original variables to participate in data modeling, so as to overcome the defects caused by too many variables in the analysis process.