2022 I/ITSEC: Multi-Modal Analytics Paper (Best Paper HPAE Subcommittee)
This paper presents a case study of teams of soldiers training on dismounted battle drills in a mixed-reality training environment. Mixed-reality simulation-based training environments along with multimodal sensing devices have made it much easier to collect and analyze participant interaction and behavior data for evaluation and feedback. Advanced AI and machine learning algorithms have further enhanced the ability to create robust multi-dimensional individual and team performance models. The performance metrics computed within single training instances, can be extended to cover a full course of training scenarios. This provides valuable feedback to trainees and their instructors on their skill levels across cognitive, metacognitive, affective, and psychomotor skill dimensions. However, developing objective data-driven performance metrics comes with a set of challenges that includes data collection and aggregation, pre-processing and alignment, data fusion, and the use of multimodal learning analytics (MMLA) algorithms to compute individual and team performance. We develop a generalized multilevel modeling framework for the training domain and use machine learning algorithms to analyze the collected training data that spans video, speech, and simulation logs. We model teams of soldiers through multiple training scenarios and show their progression over time on both operationalized domain-specific performance metrics, as well as higher-level cognitive and metacognitive processes. We conclude with a discussion of how results from our analysis framework can be used to provide formative feedback to trainees and suggestions for future training needs, as well as data-driven evidence to be used as part of a longer-term summative assessment system.