FORECASTING UNIVERSITY CONTINGENT INDICATORS USING MATHEMATICAL STATISTICAL AND NEURAL NETWORK APPROACHES

In the context of the digital transformation of education, conducting analytical research is becoming especially important to maintain a contingent of universities. This article examines the relationship between the popularity of physics-related search queries and the number of applicants choosing physics as one of the core subjects of the Unified National Testing in Kazakhstan. A neural network model for predicting the success of students completing their studies at a regional university has also been proposed and investigated. Statistical analysis and machine learning methods were used to analyze the data. The search query data was obtained using Google Trends. A regression model was built in the SPSS Statistics environment with preliminary standardization of variables. To predict the completion of students' studies, a multilayer neural network was built, trained by the Levenberg-Marquardt method. Data on the form of tuition fees, the student's place of residence, the average grade of the certificate and GPA were used as input parameters. The quality of the model was assessed using the standard deviation (MSE) and correlation coefficient (R) metrics. The developed regression model showed that 72% of the variability in the number of applicants choosing physics as a core subject is explained by the frequency of the corresponding search queries. The neural network model has demonstrated sufficient accuracy in predicting student learning outcomes. The mean square error values obtained were 0.059, 0.069, and 0.062 for the training, control, and test samples, respectively. The developed models have shown their effectiveness in analyzing and predicting student enrollment and graduation, which makes them promising for use in educational process management. However, the impact of the COVID-19 pandemic could affect the accuracy of forecasts, which requires further analysis and expansion of the database. In the future, the use of more extensive and detailed data will improve the accuracy of models and improve their applicability in the practice of educational institutions.