Técnicas de aprendizagem de máquina para classificação do nível de satisfação do estudante com base em seu perfil e percepção sobre uma universidade

Abstract

The objective of this work is to analyze the performance of machine learning algorithms to classify the level of student satisfaction, identifying the most important attributes for the models. A data set consisting of 198 high school students was used, with information about your degree, data personal experiences, routine and perceptions in the university context. From balancing techniques on the dataset like Oversampling, Undersampling and both methods combined, in conjunction with machine learning models such as Random Forest, Multilayer Perceptron, XGBoost and Support Vector Machine, the student was classified between promoter and detractor, with promoters being students who evaluate the university with grades of 8 to 10, and those detractors from 0 to 7. To evaluate the predictive power of the methods, the metrics of evaluation such as accuracy, f-score, precision and recall, mainly for the detractors class. The best model to classify students with acceptable reliability was the combination of all models in a voting system, with an overall accuracy of 90%, the most important category was grades, with 44.84% importance, and the most important variables were satisfaction with the course itself, the student’s housing situation and the student’s motivation to study.