Automation of an Educational Data Mining Model Applying Interpretable Machine Learning and Auto Machine Learning

Abstract

Educational Data Mining (EDM) is a discipline related to the development of methods that allow the extraction of knowledge from academic, socioeconomic data, and learning analytics. The information obtained is used in decision-making to improve the indexes of academic performance, educational quality, desertion and student procrastination, among others. The EDM uses techniques of Statistical Analysis, machine learning and data mining within the framework (setting) of a development methodology such as Knowledge Discovery in Databases (KDD), Cross Industry Standard Process for Data Mining (CRISP-DM), and Sample, Explore, Modify, Model, Assess (SEMMA) or variants thereof. A drawback that we have to face mainly in South America is the need to have data scientists who participate in determining stages of a conventional EDM model that applies machine learning. To minimize this inconvenience, a EDM model applied in the university environment is proposed that uses Automated Machine Learning (AutoML) and Machine Learning Interpretability (IML) techniques. These two techniques will allow us to automate stages for data pre-processing, choice of machine learning models, choice of the best hyperparameters for the models, and interpretation of the results and the characteristics that influenced in the results, these last very important from the legal and ethical point of view, because not only must the model classify or predict with a good result, but it must also be able to inform us how it reached the conclusion.KeywordsEducational data miningAutoMLMachine learning interpretability