AI-Based Student Performance Prediction System Using Machine Learning and Large Language Models

Abstract

This paper presents an AI-based Student Performance Prediction System designed to assist students in analyzing their academic performance and receiving personalized guidance. Traditional academic systems provide marks and grades but lack intelligent mechanisms to interpret data and suggest corrective actions. The proposed system integrates machine learning (ML) for risk prediction, Large Language Models (LLM) via Ollama (LLaMA 3) for conversational intelligence, and Optical Character Recognition (OCR) using EasyOCR for image-based result analysis. The system is built using a modern technology stack comprising Python, FastAPI, Spring Boot, and React.js. Experimental results demonstrate that the ML model achieves approximately 88% accuracy in predicting student risk levels based on parameters such as attendance, GPA, internal marks, and backlogs. The system also generates structured 7-day and 30-day personalized study plans and supports a natural chatbot-based interface. This work bridges the gap between raw academic data and actionable insights, offering a scalable and interactive platform for real-world academic use.