Analyzing Novice Programmers’ EEG Signals using Unsupervised Algorithms

Abstract

Ten (10) first year college programming students participated in the study and re- ported their emotions during the learning session. Emotiv EPOC headset was used to gather EEG brainwave signals. Digital signal processing filtering technique was used to filter the data. The reported academic emotions were engaged, confused, frustration and boredom. A square SOM map with 10 rows by 10 columns was built to visualize the EEG data set, a total of 100 nodes. The weights of the final SOM nodes were clustered using k-medoids and k-means algo- rithms, both derived two main clusters; one cluster aptly named “State of hope and enthusiasm” because it is primarily composed of clusters of confused emotion nodes surrounded by a topo- graphical arrangement of engaged emotion nodes; the other cluster named “State of frustration and boredom” because it is primarily composed of frustrated and boredom emotion nodes. These observations of the topographical arrangements of the SOM nodes and its subsequent clustering of the SOM nodes by k-medoids and k-means, seem to be in accordance with previous findings by (Kort, Reilly & Picard, 2001; D’Mello & Graesser, 2011) ultimately making SOM to be a viable and good alternative representation/visualization tool for D’Mello’s theory of academic affect transition model. We also observed that k-medoids required much lesser num- ber of k to derive similar clusters of SOM nodes as k-means, moreover, execution time for k- medoids is the same as k-means, making k-medoids a very attractive option for clustering algo- rithm of choice for clustering of SOM nodes.