Proceedings of the Annual Meeting of the Cognitive Science Society

With the surge in cognitive gaming data, understanding players' learning patterns and cognitive growth has become increasingly important. These data offer valuable opportunities to study individual cognitive development during learning. However, the diversity of player profiles and the complexity of gaming tasks pose significant challenges for accurate skill prediction. Specifically, the heterogeneity of player profiles leads to diverse and complex learning trajectories; data sparsity and temporal dynamics further exacerbate these challenges. To address these challenges, we propose the MFCGT (Multi-feature Forget Cognitive Game Tracing) model. First, we perform multi-feature selection to extract key features from player behavior data to reduce noise and improve prediction accuracy. Second, we introduce a time-aware decay mechanism that simulates skill degradation using an exponential decay function, ensuring the model captures the impact of forgetting on learning trajectories. Finally, we incorporate an attention mechanism to dynamically identify the most relevant historical performance for the current task, thereby enhancing the model's predictive capability. The experiment results show that MFCGT significantly outperforms traditional models in skill prediction tasks. Additionally, MFCGT effectively captures players' learning dynamics and forgetting effects, providing more accurate learning predictions.