Analysis of Student Academic Performance Based on the Data of their Midterm Examination Using Machine Learning Methods

In the modern world, education is divided into several types: in addition to the classical type of training, which is full-time and part-time, online learning is actively developing. Online, you can study both individual topics in small blocks, and receive a full-fledged education for several years. Because of COVID-19, online learning has become widespread and relevant, as in many cases around the world, universities are required to implement distance learning processes that are almost always associated with online activities. Despite the fact that the pandemic is gradually ending, online education continues to be popular and over time its popularity will only increase. It is necessary to maintain the quality of such education so that it is no worse than standard offline education. To do this, it is necessary to understand the future behavior of students in advance in order to be able to influence it. The paper explores data on the Chinese online-platform "Junyi Academy", on the basis of which models are built to predict the final grade and predict the probable outflow of a student. The novelty of the work is the ability to predict the final estimate and churn for a short period of time, as well as predicting the churn taking into account the time dependence for each solved problem.