AI-personalized programming tasks based on digital trace analysis a pilot study with pre-service computer science teachers in a containerized lab environment

Abstract

Pre-service computer science teacher education must simultaneously develop programming competence and pedagogical content knowledge, yet no empirical study has evaluated whether AI-driven task personalization based on digital trace analysis can support this dual objective in a containerized laboratory environment. This paper reports a mixed-methods quasi-experimental pilot study involving 32 pre-service informatics teacher candidates across two university cohorts in Kazakhstan. Participants in the experimental group (n = 17) received AI-personalized programming tasks generated from real-time analysis of their behavioral digital footprints within a Docker-based virtual laboratory; the control group (n = 15) completed the same tasks in the same environment without personalization. Over eight weeks, the experimental group achieved statistically significantly higher gains on both a programming skills assessment (d = 0.71, p = 0.012) and a validated TPACK survey instrument (d = 0.58, p = 0.031), with particularly strong effects on the Pedagogical Content Knowledge subscale (d = 0.84). Qualitative analysis of post-study interviews revealed that participants valued the transparency of task recommendations but expressed concern about algorithmic fairness and the accuracy of the system's assessment of their pedagogical knowledge. These findings provide the first empirical evidence that TPACK-aligned AI personalization from digital traces is feasible and effective for pre-service informatics teacher training, while identifying concrete design improvements for subsequent iterations.