Predictive Model of NAAC Results of Accreditation Based On Machine Learning Methods Using A Multi-Year Institutional Data

Abstract

India's higher education sector relies on NAAC accreditation for quality assurance, using a seven-tier grading system (A++ to C) that influences funding and autonomy. However, manual processes—lengthy Self-Study Reports, peer visits, and 12–24-month cycles—create backlogs for ~50,000 institutions. This study develops a machine learning framework to predict NAAC grades, integrating multi-year AISHE data (2010–2022) with NAAC records to form a 604,053-instance dataset. After preprocessing (KNN imputation, normalization, multicollinearity removal) and engineering 50 NAAC-aligned features, ensemble models (Random Forest, XGBoost, LightGBM) were compared. LightGBM achieved superior performance (92.3% accuracy, 91.8% F1-macro). SHAP analysis identified PhD ratio, pupil-teacher ratio, and research funding as key predictors. The reproducible model enables proactive institutional improvement and scalable accreditation forecasting.