Regresar

Unsupervised Automatic Detection of Learners’ Programming Behavior

Abstract:

Programming became one of the most demanded professional skills. This reality is driving practitioners to search out better approaches for figuring out how to code and how to support learning programming processes. Prior works have focused on discovering, identifying, and characterizing learning programming patterns that better relate to success. Researchers propose qualitative and supervised analytic methods based on trace data from coding tasks. However, these methods are limited for automatically identifying students in difficulties without human-intervention support. The main goal of this paper is to introduce a three-phase process and a case study in which unsupervised clustering techniques are used for automatically identifying learners’ programming behavior. The case study takes place in a Shell programming course in which we analyzed data from 100 students to extract learners’ behavioral trajectories that positively correlate with success. As a result, we identified: (1) a list of features that improve the quality of the automatic learners’ profiles identification process, and (2) some students’ behavioral trajectories correlated with their performance at the final exam.