ICFAI faculty team develops AI/ML architecture for granular assessment of learning outcomes
Hyderabad, Sept 15 (TNT): A faculty team from ICFAI has developed an artificial intelligence and machine learning-based architecture to assess learning outcomes at the individual student and question levels, in an approach aimed at making Outcome-Based Education (OBE) more student-centric.
The research, titled “Exploring the Deployment of AI/ML Architecture for Granular Computation of Learning Outcomes in a Foundational Online MBA Course,” was presented by Prof Prasad R at the International Conference on Educational Technology and Online Learning (ICETOL 2026) in Bremen, Germany, held from August 17 to 20.
The research was jointly developed by Rakesh P, Sunitha U L and Prasad R, the Institute said in a release here on Tuesday.
The study was conducted across five MBA courses, with 70 students in each course. It covered 927 questions across six question and assessment types, including formative and terminal assessments.
The architecture maps course outcomes and Bloom’s Taxonomy levels at the question level and computes learning outcomes for individual students, questions, question types, evaluation types and the overall course.
The study found that attainment of the five course outcomes at the overall batch level ranged between 56 per cent and 63 per cent.
Attainment across the six Bloom’s Taxonomy levels ranged between 43 per cent and 62 per cent, with comparatively lower attainment recorded at the evaluating and creating levels.
The researchers said the granular assessment could help faculty identify gaps between intended learning outcomes and student performance and assess the effectiveness of teaching and assessment methods.
The system also enables students to view their performance across different question types and rubric elements, potentially allowing more targeted feedback and self-directed learning, the researchers said.
Different AI/ML approaches were used depending on the assessment. GPT-based models were deployed for multiple-choice questions, a course-configured Custom GPT for long-form descriptive responses and a multi-call agentic pipeline for project evaluation.
The outputs were reviewed and validated by faculty.
The researchers estimated that manually computing granular OBE outcomes could take about 152 hours for a single course after the initial learning curve.
They said automation could reduce the time required for such analysis while retaining faculty involvement in system design and validation.
The team said the architecture could be extended across institutions and provide outcome-related information to students, faculty and academic administrators according to their respective requirements.
The researchers said the broader objective was to use learning-outcome data not only for accreditation and reporting but also for feedback, course redesign and quality improvement.
TNT TS
