Novice programmers’ emotion and competency assessments using machine learning on physiological data / Fatima Jannat

Fatima, Jannat (2022) Novice programmers’ emotion and competency assessments using machine learning on physiological data / Fatima Jannat. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (214Kb)
	PDF (Thesis M.A.) Download (1016Kb)

Abstract

The technology of psycho-physiological measurement and Eye-tracking has opened up a wide range of possibilities for automating the prediction of human emotional state for a particular event. There is also growing interest in modeling machine learning and deep learning algorithms that can learn from user’s data, understand and react to that individual’s affective state. This research work has used novice programming learners’ eye-tracking and Galvanic Skin Response (GSR) data in a novel approach. This work investigates the suitability and effectiveness of machine learning algorithms such as Multinomial Naive Bayes, KNN, Logistic Regression, Decision Tree for predicting levels of arousal intensity among the programmers and LSTM deep learning algorithm to classify the programmers according to their performance. Through experiments with the data-set, it was found that Multinomial Naive Bayes outperformed other supervised machine learning algorithms with 75.93% accuracy and 96.54% ROC while predicting levels of arousal intensity. Hyper-parameter tuning has been used in all the algorithms using k-fold cross validation to have the best accuracy and to avoid the over-fitting issue. The result implies a good connection between how a novice programmer goes through a programming problem and his/her emotional arousal at that moment. The Long Short-term Memory (LSTM) deep learning model was chosen for classifying programming learners according to their performance. LSTM model has the advantage of having internal memory suitable for longer sequences like our Eye-tracking and GSR data sequence. The LSTM model resulted in 65.71% test accuracy while classifying the students’ performance.

Actions (For repository staff only : Login required)