Towards Personalized Learning Environments: Using Machine Learning to Predict Students' Learning Preferences in a Mixed Reality Environment

This research-to-practice full paper describes the machine learning techniques for predicting students' learning preferences and cognitive load in a mixed reality environment. As the construction industry increasingly adopts sensing technologies, the demand for hands-on learning experiences in construction education becomes imperative. However, challenges arise in educating the future workforce due to limited access to construction sites for practical learning and the high costs associated with these technologies. To overcome these obstacles, educators are turning to virtual learning environments, such as mixed reality (MR), to provide engaging learning experiences. While MR has been utilized in construction education for simulating activities, the design of such learning environments often neglects to account for the varying learning preferences necessary for effective navigation and optimal learning outcomes. This makes it necessary to develop intelligent learning environments that can detect students' learning preferences and present information based on individual preferences. Extant studies have shown that eye movement data, particularly fixations from virtual learning environments, provide valuable insights into users' cognitive processes. Studies have also shown that leveraging machine learning algorithms enables the analysis and interpretation of eye-tracking data, offering deeper insights into students' learning preferences. In this study, nineteen undergraduate students participated in hands-on activities involving the implementation of sensing technologies within an MR learning environment. Questionnaires on learning preferences were used to assess individual learning preferences. By employing machine learning techniques on eye-tracking data and subjective evaluations of learning preferences, the study presents models that can detect students' learning preferences. The study compared performance metrics from three classifiers - Ensemble, Neural Network, and K-Nearest Neighbors (KNN) and the Ensemble classifier was identified as the most accurate, achieving an accuracy of 83.3% for the test model. The findings highlight the potential of machine learning models in detecting learning preferences most effectively for the user's interactions. The classification models, when implemented in the MR environment, can identify users who may require additional support tailored to their specific learning preferences. Thereby enhancing their interaction with the MR learning environment. Understanding and accommodating diverse learning preferences in MR learning environments can offer personalized experiences for effective engagement and knowledge retention in construction education. By tailoring instructional methods to individual preferences, educators can create inclusive environments that foster deeper comprehension and maximize learning outcomes.