KANProKT: Utilizing Kolmogorov-Arnold Networks for Superior Knowledge Tracing

In the realm of smart education, knowledge tracing (KT) plays a crucial role in enhancing students’ learning abilities by predicting future performance based on past achievements. Although many existing deep knowledge tracing models have outperformed traditional Bayesian knowledge tracing models,they limitedly employ advanced neural networks to explore higher-dimensional student interaction data. Consequently, these models often suffer from limited accuracy and interpretability. Leveraging the superior in terpretability and accuracy of Kolmogorov Arnold Networks (KAN), we introduce the KANProKT model. This novel model is the first to utilize KAN to improve performance and interpretability in knowledge tracing. We conducted experiments on three real-world public datasets, and the results demonstrate that our KANProKT model outperforms existing KT models in terms of performance and exhibits significantly better interpretability.