Impact of Pedagogical Approaches on Machine Learning Concept Mastery: A Pretest-Posttest Analysis

Abstract

The need to rapidly incorporate data science and machine learning into undergraduate math programs necessitates the development of effective pedagogical approaches to reconcile theoretical knowledge and computational skills. This quasi-experimental pretest-posttest study is a one-group, one-group design, which assessed the effectiveness of a 16-week blended instructional intervention on 94 undergraduate mathematics students of the COMSATS University Islamabad, Vehari Campus, Pakistan. The course took the form of 3(2,1) of two hours of theoretical lectures and one three-hour lab session each week, covering the basics of Data Science, Python programming and the concepts of Machine Learning. Learning outcomes in the three domains were assessed by self-constructed pre- and post-tests. Paired-samples t-tests revealed statistically significant improvements: Data Science (M_diff = 3.98, t(93) = 16.22, p < .001, Cohen’s d = 1.67), Python (M_diff = 6.59, t(93) = 21.44, p < .001, d = 2.21), and Machine Learning (M_diff = 6.64, t(93) = 23.88, p < .001, d = 2.46). Increases were demonstrated by more than 94 percent of students with big practical effects. Results indicate that blended theoretical-laboratory teaching can be successfully used in resources-limited environments and that experiential learning can be useful to teach computational mastery. Some implications on curriculum reform in higher education in developing countries are discussed.