AI-Based Knowledge Gap Detection System for Student Performance Analysis Using Predictive Analytical Models

Abstract

A single aggregate exam score hides which topics a student has actually mastered: two students with the same overall percentage may have entirely different strengths and weaknesses. This paper presents an AI-based knowledge gap detection system that analyzes quiz performance at the topic level. Each attempt is linked to a student's roll number and scored automatically, and responses are grouped by topic to compute topic-wise accuracy. Topics below a mastery threshold are flagged as knowledge gaps, and the weakest flagged topic is recommended for practice next. A separate communication module captures a spoken answer through the browser's speech-recognition interface and scores the transcript against a subject-specific keyword bank, adding an indicator of verbal articulation. The system uses a FastAPI backend, a browser-based dashboard and a scikit-learn regression baseline for accuracy prediction, and is evaluated through functional verification, analysis of a 100-student cohort data set and a held-out predictive-modelling experiment. The results show that topic-level analysis exposes gaps that an overall score conceals, while the minimal predictive baseline proves inadequate, which motivates the richer features and adaptive feedback loop proposed as future work.