New Study Uses Knowledge Components to Predict Programming Assignment Difficulty
More KCs per assignment = lower student performance, says research on 3 datasets.
A new study published at IEEE FIE 2026 proposes an interpretable framework for analyzing programming assignment difficulty using Knowledge Components (KCs). Led by Tsvetomila Mihaylova and colleagues, the research addresses a key challenge in computer science education: predicting student struggle without opaque black-box models. KCs are fine-grained concepts or skills, like 'loop nesting' or 'array indexing,' that together define what a student must know to complete an assignment. The team analyzed data from three introductory programming courses, measuring the number of KCs per assignment and changes in KC coverage between consecutive assignments.
Results showed a clear correlation: assignments requiring more KCs were associated with lower student performance. Additionally, sudden jumps in KC requirements between assignments often coincided with learning disruptions. Interestingly, some assignments saw performance declines even when no new KCs were introduced, suggesting issues in task design or instruction rather than cognitive overload. The framework allows instructors to use either expert-defined KCs or those extracted by an LLM, making it accessible. This interpretable approach provides actionable insights—helping educators identify problematic assignments, adjust pacing, and improve course design without relying on hard-to-interpret behavioral models.
- Assignments with more KCs correlate with lower student performance across three datasets
- Sudden shifts in required KCs between assignments coincide with learning disruptions
- Performance declines without new KCs indicate potential design or instruction flaws, verified via qualitative analysis
Why It Matters
Empowers instructors to pinpoint assignment design flaws using interpretable metrics, improving CS education.