Abstract
This study aimed to develop an approach to predicting academic performance based on combining cognitive classification of tasks and quantitative analysis methods. Bloom's taxonomy was used as a theoretical base, which offers a hierarchy of levels of cognitive activity - from basic operations such as memorization to more complex forms of thinking, including analysis, assessment and creative transformation of knowledge. This kind of structure allows you to more accurately assess the nature of educational tasks and their compliance with the levels of cognitive complexity. During the experiment, various algorithmic approaches were tested, among which classification methods based on decision trees, ensemble methods and models based on multi-layer data processing showed more optimal results. A significant improvement in prediction accuracy is achieved by incorporating features that reflect the cognitive level of tasks. As a result, the accuracy of the models increased from 73.4% to 91%, and the overall F1 score reached 0.86. Learning outcomes were most strongly influenced by tasks that required higher-order operations such as analysis and creation, but more accurate predictions were obtained when working with baseline tasks. The proposed approach can be used to build personalized learning systems, conduct cognitive diagnostics and adaptive knowledge assessment.