Afet Akademi — Adaptive Disaster Education System
Unity 6.3 LTS serious game with 13-model BKT, Socratic AI mentoring, and real-time learning analytics for children aged 8–16
Problem
Traditional disaster education relies on static instruction and declarative knowledge, which fails to model real-time decision-making under uncertainty. There is a need for immersive systems that can measure hesitation, contextual transfer, and adaptive behavioral calibration in high-stakes environments.
Approach
Afet Akademi integrates immersive simulation with a 13-model Bayesian Knowledge Tracing (BKT) architecture and stealth assessment principles. The system models learner knowledge states in real time across 10 specialist models — including MetaCognition, Confidence, ErrorPattern, and HelpSeeking — calibrated per Piaget's developmental stages for three age groups (8–10, 11–13, 14–16). A Socratic AI mentor (BİLGE) guides reflective thinking without providing direct answers. The companion web platform (afetakademi.com.tr) collects anonymized behavioral data across 4 measurement layers for academic analysis.
Outcome
The system demonstrates a structured framework for modeling behavioral calibration in disaster preparedness education. By embedding assessment invisibly within gameplay, Afet Akademi shifts evaluation from right/wrong scoring to cognitive state inference. The project establishes a foundation for adaptive immersive learning infrastructures and explainable learning analytics integration. All behavioral data is exportable as CSV for SPSS, R, and Python analysis.
Technology Stack
Measurement Architecture
What the system actually measures across 4 layers
Layer 1 — Identity & Session
Anonymous session_id (KVKK compliant, no personal data), device type, timestamp
Layer 2 — Game Events
game_start — game started · game_action — every decision/move · game_end — game finished
Layer 3 — Performance Metrics
Score (0–100), time spent, completion status, objectives completed, accuracy rate
Layer 4 — Behavioral Patterns
Response time, fast-but-wrong pattern (< 2s + < 50% accuracy), decision sequence, resource efficiency
Research Question
Outcome × Game Type Matrix — which game type (quiz, strategy, 72-hour) best measures/teaches which learning outcome?
All data exportable as CSV — compatible with SPSS, R, and Python for academic analysis.
afetakademi.com.tr
Learning Analytics & XAI Platform
Interested in similar projects?
Discuss a Project →