Early prediction of student performance in CS1 programming courses

There is high failure and low academic performance in programming courses. To mitigate these problems it is necessary to predict student performance at an early stage. This will allow teachers to support students with interventions that allow them to achieve the defined learning objectives. Student performance prediction is gaining more and more attention, researchers are working on machine learning features and algorithms that can be used to achieve better results in predictions. This paper proposes a model that predicts student performance in weeks 3, 5 and 7 of a 16-week CS1 programming course.  Starting from the grade, delivery time and the number of attempts generated by the student in programming labs and an exam. Eight classification algorithms were used to train and evaluate the model. The metrics used were: accuracy, recall and F1 score. The algorithm that presented the best results in week 3 was Gradient Boosting Classifier (GBC) with an F1 score of 86%, followed by Random Forest Classifier (RFC) with 83%.