Use of Machine Learning to Identify Predictors of Student Performance in Writing Viable Computer Programs with Repetition Loops and Methods

The goal of this research to practice full paper is to identify which computer programming concepts/skills predict students' ability to write viable programs using repetition loops and custom methods in Java. We developed machine learning models (logistic regression and decision trees with Scikit-Learn) to predict student performance in writing computer programs. High scores in feature importance analysis of the concepts/skills used as inputs to the models were considered important for predicting the output (i.e., performance in writing viable programs using loops and methods). We found that: 1) The ability to write programs with repetition and methods relies on an adequate understanding of several previous pre-requisite concepts/skills; 2) The relative importance of the pre-requisite concepts/skills varies, but adequate understanding of selection structures, which is typically taught early in the semester, is critical for students to be able to write viable computer programs using repetition loops and methods later in the semester; 3) Machine learning models can be used as predictors of student ability to write viable computer programs; 4) The transparency and interpretability of white-box models, like logistic regression and decision trees, allows students and teachers to identify which pre-requisite concepts/skills need to be emphasized and reinforced to increase performance on a target concept/skill.