Learning Analytics Should Analyse the Learning: Proposing a Generic Stealth Assessment Tool

Abstract

Stealth assessment could radically extend the scope and impact of learning analytics. Stealth assessment refers to the unobtrusive assessment of learners by exploiting emerging data from their digital traces in electronic learning environments through machine learning technologies. So far, stealth assessment has been studied extensively in serious games, but has not been widely applied, as it is a laborious and complex methodology for which no support tools are available. This study proposes a generic tool for the arrangement of stealth assessment to remove its current limitations and pave the road for its wider adoption. It describes the conceptual design of such a tool including its requirements regarding users, functions, and workflow. A prototype was implemented as a basic console application covering the tool's core requirements, including a Gaussian Naïve Bayes Network utility. Generated input files were used for testing and validating the approach. In a controlled test condition the stealth assessment classification accuracy was found to be inherently stable and high (typically above 92%). It is argued that the proposed approach could radically increase the applicability of stealth assessment in serious games and inform current learning analytics approaches with unobtrusive, more detailed and genuine assessments of learning.