An Intelligent Educational Assessment and Automated Question Paper Generation Framework Using Machine Learning and Generative AI

Creating well-balanced question papers, difficulty control, marking and analysing students' performance are mostly manual and non-integrated in most institutions. Previous work has focused on each of these components of this workflow, with the automatic question generation or performance prediction components being studied separately; few studies have explored them together in a system. This paper proposes an integrated framework that combines (i) Approved Syllabus Content-based Generative AI Question Generation (GAIQG) with teacher's review, (ii) Structured question bank with subject-wise, topic-wise and difficulty-wise organisation, (iii) Blueprint-driven automatic question-paper generation formulated as a constrained selection problem, (iv) Automatic assessment and topic-wise analytics, and (v) Machine Learning models for learning-level classification and adaptive testing based on Item Response Theory. The architecture is explained as a nested structure including access control based on roles and a teacher-in-the-loop gate. As a feasibility check for the analytics components, we run two experiments on synthetic data. A SVM (RBF) classifier achieves 78.0% on test set and macro-F1 of 0.786 in three-level learning classification on simulated cohort of 1500 students. The ability-estimation error (RMSE 0.40) is the same with adaptive item selection as it is with 29 items (62% fewer). The results obtained are preliminary and will be validated with real institutional data in the following phase.