F9-SP2-1 - Data Mining Approach for Determining Student Attention Pattern
3. Research Full PaperM. Jean Mohammadi-Aragh1, John E. Ball1
1 Department of Electrical and Computer Engineering, Mississippi State University
Introduction
Abstract—This Research Full Paper presents the approach of traditional engineering analysis techniques on education. Data mining techniques have been successfully employed to extract hidden information from large data sets within various contexts. We hypothesized that data mining techniques can similarly be applied to large educational data sets to extract and analyze patterns and create insights. Specifically, we examined the degree to which standard data mining techniques can distinguish between different student attention patterns in large lectures in which personal computers were actively used. Our data set consists of electronically captured student attention data (on-task, off-task) that was recorded at 20 second intervals throughout each course lecture over one semester. With Institutional Review Board (IRB) approval, methods involved capturing student data via a backend monitoring system to reduce student awareness of monitoring and reduce false behavior changes during data collection periods. The data were originally captured in the form of images (screenshots), and image processing techniques were applied to extract student attention patterns in the form of zero (off-task) and one (on-task). We conducted descriptive statistical analysis to add other features such as characterization information (e.g., total logged in attention, average class period attention to the recorded data sets). For the data mining analysis, we used three different supervised machine learning classification algorithms: Support Vector Machine (SVM), Decision Tree, and K-Nearest Neighbors (KNN) for classifying the students’ dataset. We classified the students into one of four different classes based on their attention pattern in the lecture class. Before applying each classification algorithm, feature extraction was performed. For this purpose, we used Haar wavelets, Principal Component Analysis (PCA), and Linear Discriminant Analysis (LDA) dimensionality-reduction techniques before applying the classification algorithm. Their performance is compared. Our results indicate high classification accuracies can be obtained using these dimensional reduction algorithms followed by classification algorithms. Our result highlights the importance of applying traditional engineering analysis techniques to educational data in order to provide engineering education insights.
Keywords— Data mining, SVM, Decision tree, KNN, learning analytics