SARS: An Explainable Student Academic Performance Risk Assessment and Advisory System

Abstract

Academic institutions worldwide collect detailed student performance data across multiple semesters, yet existing academic information systems serve primarily administrative purposes without providing interpretive, predictive, or actionable insights. Faculty advisors struggle to identify at-risk students early enough for effective intervention, while students lack tools to understand their academic trajectory and receive personalized guidance. Predictive systems that do exist frequently operate as black boxes, generating opaque risk scores that limit practical utility and raise transparency concerns in educational settings where trust and student agency are paramount. This paper presents SARS — a Student Academic Performance Risk Assessment and Advisory System — an explainable decisionsupport platform designed for early academic risk detection and personalized advisory support in higher education institutions. The system makes three principal contributions: (1) an automated grade ingestion pipeline using a multimodal document extraction workflow to convert photographs or scans of official grade sheet documents into structured academic records; (2) a multi-factor risk scoring engine computing a SARS Score from 0 to 100 as a credits-weighted composite of GPA trend (40%), backlog accumulation (35%), and attendance compliance (25%) with placement policy floor rules; and (3) an evidence-grounded advisory pipeline using ChromaDB vector storage and Gemini text-embedding-004, where student data is decomposed into typed knowledge chunks retrieved semantically per query with source citations accompanying every response. The system was validated using real grade sheet documents spanning five semesters and demonstrated across a synthetic cohort covering all four risk levels. Evaluation confirms formula accuracy, risk classification alignment with institutional policy, and effective retrieval across varied advisory question types.