Evolving Classification of Learners' Profiles using Machine Learning Techniques for Better Retention Rates in Massive Open Online Courses

High dropout rates and low completion rates have been associated with Massive Open Online Courses (MOOCs) ever since its advent. The rigid structure, lack of feedback and interactivity provided by MOOCs imparts students with lack of motivation to complete the course. While many studies have been formulated to offer solutions to these problems, most of them require major design changes and are not cost effective. This paper aims to provide a design that can be used universally with any course and doesn't require design change in already existing MOOCs. Objective: This study aims to build evolving classifier models, thatd classify students constantly based on their interactions with the courses materials. Dataset: The dataset is obtained from Open University Learning Analytics, which contains details of students interaction with several courses materials and other demographic details of students. Methodology: This research project compares 7 different classification algorithms, all implemented and evaluated in Python. The first occurrence of a sequence of 5-day data is trained on all the models, and the best performing model is then selected to continuously classify the students as the course progresses. Results: This study find the Area Under ROC curve (AUC) score for every classifier. A score greater than 0.70 is considered a very strong course in studies in regard to predicting future behaviour changes, XGBoost outperformed all the algorithms, with AUC score of 0.73