A MODEL FOR DETERMINING THE DIRECTIONS OF SCHOOL GRADUATES BASED ON MACHINE LEARNING ALGORITHMS

Abstract

The purpose of this article is to determine the orientation of schoolchildren in terms of grades in
subjects and soft skills using machine learning methods. The article examined the difficulties of constructing a
sigmoid function using multivariate linear regression, and also digitized grades obtained in selected subjects in the
field of student education over 10 years and their various parameters, reasons and student capabilities. . Using these
numbers, a training data set was created. As a result, a classification of subjects studied by schoolchildren over 10 years
and their assessments was developed. Neural network architectures, modules, the most commonly used activation
functions in machine learning algorithms, training methods and methods for constructing linear and logistic regression,
disadvantages and opportunities are analyzed. Ways to simplify the gradient descent function for multivariate linear
regression by vector calculation have been studied. Because there are many variables involved in this type of linear
regression, vector calculations have proven to be more convenient. Methods for parallel calculation of gradient descent
processes using vector calculations are also considered. In particular, the addition of training data table columns,
transposition of coefficients - AT, vectorized representation of a linear function, hyperparameters for gradient descent
(learning rate - , number of steps) were defined.