Board 11: Predicting At-Risk Students in a Circuit Analysis Course Using Supervised Machine Learning

Writing exercises may be used in problem-centric STEM-based courses to identify common misconceptions held by the writer as well as to probe their metacognitive processes. As grading of writing samples and providing personalized feedback regarding a student’s writing can be timeintensive, opportunities to automate the process while retaining the integrity of the grading and quality of the feedback are attractive. This paper describes the motivation and use of a writing-based exercise in a sophomore-level course on electric circuit analysis. The conversion of a paper-based writing exercise to a webbased application is detailed as is its initial use in this new format. The ultimate goal of implementing a web-based approach to administering the writing exercise is to build a fully automated application capable of evaluating student responses and providing feedback to the user in an attempt to enhance their conceptual understanding of challenging material in a manner that acknowledges instructor workload in high-enrollment, resource-constrained courses. The first element in the planned automated evaluation aspect of the writing application is the identification of students scoring at the lowest end of a holistic scale. This is of significant value as there is evidence that such students are at-risk to fail the electric circuits course as it is currently constructed. Use of a basic natural language processing (NLP) pipeline on a dataset of more than one hundred student responses is described as are the initial results of the at-risk / not at-risk binary classification task.