Struggling to Detect Struggle in Students Playing a Science Exploration Game
summary
Built a real-time struggle detector for the learning game Wake: Tales from the Aqualab using interaction logs, four engineered feature sets aggregated at the task level, and nine classification algorithms. Ground truth comes from human-labeled text-replay clips (Struggle / Not Struggle / Bad Clip). Naive Bayes performs best under student-level five-fold cross-validation with AUC = 0.635; SHAP explains feature importance. Work flags modest accuracy, definition ambiguity, and potential algorithmic bias as next targets.
AI-written summary · about
related talks
06mUnderstanding Player Patterns by Combining Knowledge-Based Data Abstraction with Interactive ...
53mBalancing gameplay of aRPG using player behavioral patterns
15mUsing Information Visualization to Promote Students' Reflection on "Gaming the System"
15mUnderstanding Learner Behavior Through Learning Design Informed Learning Analytics
21mMining Controller Inputs to Understand Gameplay
38mDeep Learning plays Handball
14mModeling Individual and Team Behavior through Spatio-temporal Analysis
05mA Data-Driven, Player-Centric Approach to Evaluating Spatial Skill Training Games