[2603.28553] Multimodal Analytics of Cybersecurity Crisis Preparation Exercises: What Predicts Success?

Computer Science > Human-Computer Interaction arXiv:2603.28553 (cs) [Submitted on 30 Mar 2026] Title:Multimodal Analytics of Cybersecurity Crisis Preparation Exercises: What Predicts Success? Authors:Conrad Borchers, Valdemar Švábenský, Sandesh K. Kafle, Kevin K. Tang, Jan Vykopal View a PDF of the paper titled Multimodal Analytics of Cybersecurity Crisis Preparation Exercises: What Predicts Success?, by Conrad Borchers and 4 other authors View PDF HTML (experimental) Abstract:Instructional alignment, the match between intended cognition and enacted activity, is central to effective instruction but hard to operationalize at scale. We examine alignment in cybersecurity simulations using multimodal traces from 23 teams (76 students) across five exercise sessions. Study 1 codes objectives and team emails with Bloom's taxonomy and models the completion of key exercise tasks with generalized linear mixed models. Alignment, defined as the discrepancy between required and enacted Bloom levels, predicts success, whereas the Bloom category alone does not predict success once discrepancy is considered. Study 2 compares predictive feature families using grouped cross-validation and l1-regularized logistic regression. Text embeddings and log features outperform Bloom-only models (AUC~0.74 and 0.71 vs. 0.55), and their combination performs best (Test AUC~0.80), with Bloom frequencies adding little. Overall, the work offers a measure of alignment for simulations and shows that multimoda...