Improving Confusion-State Classifier Model Using XGBoost and Tree-Structured Parzen Estimator

Abstract

Detecting confusion has been considered as a critical issue in online education platforms. Confusion emerged as an effect of the limited interaction between lecturers and learners. The confusion detection machine learning model can be used to overcome the problem. Such a model can provide the ability for online education systems to detect confusion, thus it can react accordingly. Encouraged by the need, several studies have been done to develop confusion-state classifier models. The best previous model has an average accuracy of 75%. Despite having a promising result, the model still contains several gaps that can be improved. The gaps lie in the selection of the machine learning algorithm and the absence of any hyper-parameter optimization technique. This study aims to overcome them using two approaches: replacing the machine learning algorithm with XGBoost and applying the Tree-structured Parzen Estimator (TPE) as a hyper-parameter optimization technique. The TPE was also combined with the Recursive Feature Elimination (RFE) technique. The proposed model had outperformed the previous ones by achieving an average accuracy of 87%. This study also brought out the most optimal configuration of features and hyper-parameters to build such a model. This study had presented the current confusion-state classifier model.