Northeast China is a mid- to high-latitude region with significant lake distribution and high sensitivity to climate change. Its unique seasonal lake ice phenology changes have a substantial impact on the lake ecosystem. However, there is a lack of long-term lake ice phenology data in Northeast China, making it difficult to identify the characteristics of lake ice phenology changes. To this end, this study employed the XGBoost-SHAP machine learning model to construct a lake ice phenology dataset for key lakes in Northeast China (Hulun, Lianhuan, Chagan, Xingkai, and Wolong Lakes) from 1981 to 2023, and conducted a quantitative analysis of the characteristics of lake ice phenology changes and their driving factors. The results show that (1) the XGBoost-based lake ice phenology prediction model achieves high accuracy. Specifically, the coefficient of determination (R2) for predicting freeze-up data is 0.97, with an average absolute percentage error of 0.5%. For predicting break-up data, the R2 value is also 0.97, but the average absolute percentage error is slightly higher at 1.9%. (2) Lake ice phenology exhibits the characteristics and trends of delayed freeze-up data, advanced break-up data, and shortened ice cover duration. At high-latitude Hulun Lake, freeze-up is delayed by 0.18 d/a, break-up advances by 0.37 d/a, and ice cover duration is shortened by 0.55 d/a. In contrast, at the relatively lower-latitude Wolong Lake, freeze-up is delayed by 0.13 d/a, break-up advances by 0.20 d/a, and ice cover duration is reduced by 0.33 d/a. (3) The primary driver of lake ice phenology changes is air temperature, contributing 40.5% and 31.2% to freeze-up and break-up data, respectively. The study"s findings enhance our understanding of the response mechanisms of lake ice phenology in cold regions to global climate change and provide scientific support for lake water environment protection and water ecology management.