Abstract:

The demand for graduate computing professionals has steadily increased over the past numbers of years with the IT sector regularly requiring 140,000 entrants each year in the UK. However, according to the Higher Education Statistics Agency there are only 16,000 computing graduates per annum, leaving a shortfall of 120,000. Consequently universities have responded by increasing their intake in computing degrees, and also a significant growth on enrolments on conversion MSc Programming courses.

High attrition and failure rates are especially prevalent in computer programming based courses, with attrition rates being reported at around 11% in the UK universities. Due to the fundamental importance of programming to CS, most CS degrees begin with introductory programming modules. However, regardless of the recognised importance of learning programming, the outcomes are often disappointing. Many institutes report dropout rates of 20-40% of students on such courses. Given the severe skills shortage in the information and computer technology (ICT) sector worldwide these high non-progression rates in CS courses are of considerable anxiety.

The prediction of outcomes for undergraduate programming modules has been attempted via aptitude tests. However, most of the approaches to date either attempt to predict long term performance, but with a relatively low level of predictability, or attempt to identify those students that are not learning but often not at an early enough stage where intervention efforts are likely to be most successful.

In contrast to undergraduate courses the increasing popular MSc Computing Conversion courses uniquely provides new pre-course data points that could be used to help predict performance. This study individually analyses separate themes including previous STEM vs non STEM degree background, previous degree classification, previous subject knowledge and aptitude test scores. It analyses several separate cohorts and reports on the ability of the separate themes to predict outcomes. It then provides several Machine Learning classification models of the combined themes.

It finds that there are several themes that provide statistically significant correlations including pre-course aptitude testing that could be used to identify applicants that are unlikely to pass a programming modules or identify those that may struggle and would benefit with additional support. The outcomes provide insights in programming aptitude and could useful help contribute solving the significant issue of high failure and dropout rates in programming courses.

Keywords:

Programming, prediction, machine learning.