Incorporating Unstructured Text in Multi-Layer Perceptron (MLP) Network: Factors Affecting Partner Selection in Pair Programming

The revealed analysis studies on pair programming so far indicate that pair programming has produced affirmative effects on some aspects of students” performance. In the academic field, the usual practice of pair programming would be pairing the students in line with the programming skills of the students by the respective lecturers. This means, compatibility of the students in terms of their programming skills is the main focus when the pairing was done by the lecturers. Yet, research on elements that the students are looking into when they are given the liberty to decide on their partner in pair programming is lacking. In this study, a multi-layer perceptron (MLP) is developed to predict the preference of opting pair programming over solo programming. The Bayesian Information Criterion was used to select the best features in the prediction. The potential of unstructured text entered by the participants as comments in the questionnaire is incorporated in the MLP model to verify its capabilities towards prediction accuracy, i.e., to verify whether their comments are connected to their preference for pair programming versus solo programming. It was found that, when the students are given the freedom to choose their partner in pair programming, in the context of Malaysia, the students would pay attention to the ethnic criterion. This also suggests that the unstructured texts in the form of comments submitted by the participants in the questionnaire did not contribute to their choices on whether to undertake solo or pair programming.