SynthEd: From Synthetic Data to Simulated Learners

SynthEd is an agent-based simulation environment for open and distance learning (ODL) research that shifts the focus from static data generation to dynamic learner simulation. By combining persona-driven agent modeling with 11 established theoretical frameworks (Tinto, Bean & Metzner, Kember, Bäulke, SDT, Garrison CoI, Moore, Rovai, Epstein & Axtell, Gonzalez, and Lazarus & Folkman), SynthEd generates behaviorally grounded and temporally coherent learning trajectories. Each synthetic student's weekly engagement, dropout risk, GPA, and social network interactions emerge from their evolving motivations, experiences, and life contexts. Key capabilities: Trait-based calibration pipeline: Sobol sensitivity analysis (51 parameters), Optuna Bayesian optimization, and held-out OULAD validation (Grade B on unseen modules) Configurable populations calibrated to institutional demographics Multi-semester simulation with inter-semester carry-over mechanics GPA feedback loop into cost-benefit, dropout phases, and competence beliefs OULAD-compatible 7-table export for drop-in EDM research compatibility Adaptive parameter bounds (auto_bounds) that adjust to user-modified defaults Optional LLM-enriched backstories with culturally diverse name generation 5-level validation suite (distributions, correlations, temporal coherence, privacy, backstory consistency) 4 benchmark institutional profiles for reproducible research 452 tests, 99% coverage Use cases: dropout prediction model development, intervention simulation, privacy-safe benchmarking, and educational data mining research in GDPR/KVKK-constrained settings.