A Mixed Methods Framework for Usability Profiling in Learning Environments: Integrating Adapted Heuristics and Machine Learning

Traditional usability evaluations of Learning Management Systems are often based on aggregate metrics that mask the heterogeneous experiences of diverse user populations, thus yielding limited actionable insights for personalized improvements. To address this significant gap, this research proposed a systematic six-phase methodological framework that integrates user-centered design principles, contextually adapted educational heuristics, and advanced machine learning techniques to deconstruct usability data and generate distinct data-driven user profiles. The methodology consisted of an empirical single-case study using a mixed-method approach to evaluate the implementation of a virtual Moodle environment for adult education in Comfacauca in Colombia. Data were collected from a heterogeneous sample of 50 students using the System Usability Scale questionnaire and the qualitative think-aloud protocol. These datasets were analyzed by combining descriptive statistics with Principal Component Analysis and K-means clustering. Although the platform achieved a strong aggregate SUS score of 73.9, analytical results enabled the successful identification of four distinct user profiles with unique technical needs, behavioral patterns, and cognitive challenges, which would have remained hidden with conventional single-score methods. For example, the framework revealed that perceptions of usability were more strongly correlated with the students' academic cycle than with their chronological age, directly challenging common demographic stereotypes. Finally, it is concluded that this comprehensive framework provides a reproducible methodological protocol for educational institutions to gain deeper, user-centered insights. This approach enables the transition from generic adjustments to the development of highly specific and differentiated support strategies, demonstrating the profound value of combining traditional usability methods with unsupervised clustering algorithms to transform general user feedback into actionable pedagogical intelligence.