This study conducts an ablation analysis on existing implementations of forgetting and question difficulty in Deep Knowledge Tracing (DKT) models. Experiments with ten state-of-the-art DKT models across five prominent datasets provide insights on the effectiveness of incorporating forgetting and question difficulty components. Results indicate that modelling forgetting as an exponential decay term within self-attention mechanisms appears to be the most effective approach. Additionally, incorporating question difficulty using the Rasch model – both in input embeddings and self-attention mechanisms – yields the best results. Furthermore, datasets with long-term student interaction histories best capture the effects of forgetting and question difficulty, aligning with recent theories on human forgetting behaviour and the influence of question difficulty on learning. These insights empower Intelligent Tutoring Systems (ITSs) to leverage student cognition while learning and tailor DKT models deployed, enhancing teaching and learning.