Machine learning to identify key success indicators

. This article explores the application of machine learning techniques in the context of identifying and analyzing key indicators of learner success. In particular, the paper focuses on the application of machine learning techniques such as decision trees, Kohonen maps and neural networks. Decision trees are a graphical model that helps to analyze and make decisions based on hierarchical data structure. They allow classification and regression analysis, which helps in highlighting optimal strategies and recommendations to improve learner success. Kohonen map are used to highlight key success indicators, find hidden patterns and group data. Neural networks are able to analyze complex relationships and predict outcomes based on input data. The selected machine learning methods allow to optimize the learning process, adapt teaching methods to individual needs and increase the effectiveness of education in general.


Introduction
Machine learning plays an important role in today's world and has a significant impact on various industries and fields.It can help in determining the effectiveness of educational programs, predicting student performance, and identifying educational needs.The approach can significantly improve the quality of education and help make more informed decisions.
In education, there is potential to use machine learning as a decision support tool, but its application is still limited.Organizations and educational institutions often do not use the available machine learning tools for data analysis, viz:  Forecasting and Prediction;  Recommender Systems;  Data Processing;  Process Automation and Optimization;  Pattern Recognition and Computer Vision [1,2].
The purpose of this paper is to obtain machine learning models that will help in decision making related to curriculum optimization, prediction of student performance, personalization of learning and other important aspects of the educational process [3].
The use of machine learning as a decision support tool will improve the quality of decisions made in the educational field.The models will contribute to more accurate prediction of student performance, curriculum adaptation and personalization of learning, which will ultimately lead to improved efficiency and effectiveness of educational institutions [4,5].

Materials and method
In modern conditions, there is a toughening struggle for intellectual resources, which are one of the main advantages and act as the basis for the welfare of the enterprise.To obtain higher education, modern youth carefully approaches the choice of educational institution and the direction that is the most interesting for him [6].Success in training forms a competitive and demanded in the labor market employee, capable of self-development [7].
Scientists and educators are still looking for new dependencies between different factors of student dropout [8].Someone believes that it is a step backward, for example, sociologist Steven Seidman believes that dropping out is an interruption of studies, interfering with a student's continuous advancement [9].Someone, on the contrary, says that dropping out gives an opportunity to find oneself and a new path in life, to discover other opportunities and become successful on one's own, without higher education in educational institutions.It is worth noting that a greater number of dropouts are due to academic failure, lack of ability to pay for education, as well as the quality and conditions of education [10,11].
It is virtually impossible to make comparisons between countries or even institutions, not only because of differences in educational regulations but also because of differences in how dropout is measured.It follows that factors affecting success at one institution may not have the same relationship at another institution.
To simplify the data, a data analysis method that automates the construction of an analytical model, also called machine learning, is used to simplify the data.The method can adapt to new data, is capable of learning, and prediction [12].Data analysis and possible interpretation of the data were performed using decision trees, Kohonen maps, and neural networks.
A decision tree is a structure that resembles a flowchart [13].This algorithm automatically selects features for nodes and constructs decision rules that are understandable to an expert.It predicts the value of the target variable based on the information obtained from the characteristic variables, which in turn measure the likely fit to a certain class [14].A selforganizing Kohonen map is a method of projecting a multidimensional space onto a twodimensional space, consisting of two layers: input and output [15].Kohonen maps allow clustering of objects by forming clusters, as well as predicting and detecting features and patterns in large data sets [16].Neural network is used for modeling nonlinear systems and allows to detect complex dependencies in data [17].

Results
While performing the study on the three methods, it was observed that the decision tree and neural network methods have large error, due to which only Kohonen maps will be considered in this paper.The error information is presented in Table 1.Before starting to work with the data, correlation analysis was conducted to identify the significance of factors, its results are presented in Table 2.After the correlation analysis, the data were processed and presented using selforganizing Kohonen maps.This visualization allows to trace the interrelation of clusters Figure 1.As a result of analyzing the maps obtained, portraits of students were drawn based on the value of the success criterion Table 3.

Attends but rarely Attends
Table 3 shows that success is determined by grade level, parental education, time spent with parents, and attendance at extracurricular activities.

Discussion
Machine learning has become a powerful tool for analyzing data.Its main advantage is its ability to find patterns in different complex data sets.Traditional data analysis methods are often based on subjective judgments or predetermined metrics that do not fully reflect the true essence [18,19].
Machine learning models are changeable, i.e., they can be retrained to make changes to analyze subsequent data [20,21].In order for a model to be properly trained, it is necessary that the data under study be carefully processed.It is also worth noting that for individual datasets from different localities or educational institutions, the models needed to analyze them cannot be universal.A separate decision-making model is built for each individual location [22,23].

Conclusion
The main question of this study is to identify the differences between successful and unsuccessful students.The criterion was defined only by the cumulative value of subject grades, which brings its own peculiarities to this study.When considering the dataset, those factors that had a greater dependence on the output parameter were taken into account without also considering obvious attributes such as subject grades and number of absences.
Kohonen maps were used to identify factors influencing the success criterion.Such factors were: parental education, time with parents, and extracurricular activities [24].
The resulting portraits of students based on the analysis of Kohonen maps can help teachers to assess the student's capabilities at the beginning of the discipline.There are also many other factors that influence student success.Based on Kohonen chart analysis one should: 1. Encourage parents to improve their education; 2. Spend more time with parents both in terms of checking homework and helping them to complete it; 3. Involve students in extracurricular activities [25].

Table 1 .
Uncertainty in model training methods.

Table 2 .
Correlation analysis of the data.

Table 3 .
Portrait of a student.