Financial crises have become more frequent in recent years, such as the Great Depression of 1930 and the global financial crisis of 2008–2009. Predicting financial crises is particularly important, so this paper focuses on using econometric models and machine learning to predict the possibility of financial crises. Existing prediction models are limited in variable selection, so this paper uses a wider range of economic indicators. In terms of research methods, this paper preprocessed the data and used techniques such as time series analysis, lagged variable regression, logistic regression, and multicollinearity testing to identify and verify critical variables. The results show that credit, loan growth rate, and loan-to-credit ratio significantly affect the occurrence of financial crises, while real GDP growth can reduce systemic risks. First, this study provides policymakers with a more refined perspective on formulating policies and provides a scientific basis for macroeconomic policy formulation and risk prevention. Second, this study also improves the original model, which helps predict financial crises more accurately.