Analyzing Curricular Pattern Complexity Using AI to Improve On-Time Graduation Rates

Research Objective and Problem Context

• Undergraduate curricula—especially in Software Engineering—often contain long prerequisite chains where failure in a single course jeopardizes timely degree completion within four years.
• Manual curriculum analysis and revision by faculty is time-consuming and infrequent, hindering responsiveness to evolving student needs and industry demands.

Methodology and Technical Approach

• Based on arXiv:2607.13094v1, the approach leverages Large Language Models (LLMs) to model and quantify curricular pattern complexity—including prerequisite dependencies, credit distribution, semester workload, and failure propagation paths.
• LLMs identify structural bottlenecks (e.g., high-failure-rate gateway courses, cross-semester tightly coupled modules) and generate pedagogically valid, degree-compliant revisions (e.g., splitting high-risk courses, inserting buffer modules, relaxing prerequisite constraints).

Empirical Impact and Value

• Validated on a real-world Software Engineering undergraduate program: curriculum revision cycle reduced from 6–12 months to 2–3 weeks.
• Simulation predicts a 28.4% reduction in critical-path failure probability and an estimated 5.7–9.2 percentage-point increase in four-year on-time graduation rate.
• The method avoids custom rule engines; instead, it relies on prompt engineering and structured output constraints to endow LLMs with interpretable, auditable curriculum reasoning capabilities.