Towards Final Scores Prediction over Clickstream Using Machine Learning Methods

Abstract

E-books are capable of producing a significant amount of clickstream data thatinsights students’ learning behavior. Clickstream data are often analyzed in learninganalytics and educational data mining domains to understand students’ synchronous andasynchronous learning processes. The present study analyzed a dataset consisting ofuniversity students’ clickstream data for predicting their final scores using machine-learningmethods. To begin with, the raw data are preprocessed in four steps, namely dataaggregation, feature generation, data balancing, and feature selection. After that, utilizingmachine learning methods, high performing and low performing students’ final scores arepredicted. For this, eight machine-learning methods (Neural Network, AdaBoost, LogisticRegression; Naïve Bayes, kNN, Support Vector Machine, Random Forest, and CN2 RuleInduction) are employed and their performances were compared. Result revealed that CN2Rule Induction algorithm having 88% accuracy outperformed other machine learningmethods when best-5 selected features from the dataset were taken into consideration.However, the Multilayer Perceptron based Neural Network performed best having thesimilar accuracy with CN2 Rule Induction when all features were considered to predict.This paper also focuses on how SMOTE as a data balancing algorithm can be applied tosolve data imbalance problem and various scoring methods can be compared to identify themost important feature attributes in clickstream.