To address the lagging nature of traditional academic early-warning systems and the homogenization of support interventions， this study innovatively integrates ensemble learning with causal inference algorithms. An AdaBoost model， validated on data from 2 003 students， was employed to achieve precise stratification and real-time monitoring of academic risk. A causal forest model was then used to quantify the net effects of interventions， enabling the development of multidimensional personalized support strategies.Three types of higher education institutions at different tiers

were selected， with each type including 50 students in the experimental group （AI-based early warning and personalized support） and 50 students in the control group （traditional early warning and conventional support）. Following two academic years of intervention and longitudinal tracking， the results showed that the experimental group achieved an average GPA increase of 35.70%， significantly higher than the 19.40% observed in the control group. Improvements in learning behavior compliance rates and academic adaptability scale scores were also significantly greater in the experimental group （

<math id="M2"><mi>P</mi><mo><</mo><mn mathvariant="normal">0.05</mn></math>

https://html.publish.founderss.cn/rc-pub/api/common/picture?pictureId=108702270&type=

https://html.publish.founderss.cn/rc-pub/api/common/picture?pictureId=108702246&type=

13.29266548

2.70933342

）.The findings demonstrate that AI-based early-warning models validated on large-scale datasets can accurately identify underlying academic risk factors， while personalized support enables targeted interventions. The synergistic effect of these two approaches produces a significant positive impact on the academic performance of at-risk students， providing a replicable framework for the development of academic support systems across higher education institutions of varying tiers.