Faktori učenja i predviđanje uspešnosti u programiranju primenom veštačkih neuronskih mreža

Abstract

Academic education is one of the key areas in the process of modernization of a country. The ability to predict success helps teachers identify students who have the potential to attend advanced courses, as well as students who need additional education. In modern society programming skills are becoming increasingly important. Many studies show that programming is one of the critical skills of students' technological literacy. Therefore, there is a need to analyze a large amount of data on the basis of which factors that affect student performance in the field of programming can be predicted. In recent years, the application of artificial intelligence in education has increased significantly worldwide. Artificial neural networks (ANN), as one of its tools, are experiencing numerous successful implementations. In the doctoral dissertation Factors of learning and predicting success in programming using artificial neural networks, the ANN model developed for the purpose of predicting the success of students in acquiring programming knowledge and skills is presented. 180 students of the study program Information Technology from the Faculty of Technical Sciences in Čačak were analyzed. Data on previous education were collected for each student. Students' success in learning programming is measured through achievements on the knowledge test and is classified into three categories: unsuccessful, moderately successful and very successful. A three-layer ANN model based on a backpropagation learning algorithm was used to predict student success. 19 models were created. The model with the best predictive accuracy (90,7%) was used as the final model for implementation. A web application was created for that model, with the help of which the teacher has the possibility of adapting the teaching, and more efficient organization of the same, which leads to successfully mastered material.