Accurate Forecasting of Teacher Performance and Behaviour in Higher Education Using RNN-XGBoost Multilayer Model

Precisely predicting teacher performance and conduct in higher education is an increasing difficulty owing to the complex and variety of human-centric data. Current methodologies frequently depend on static attributes and do not adequately account for the temporal dynamics crucial for prediction precision. This study introduces a hybrid forecasting framework that integrates Recurrent Neural Networks (RNN) with eXtreme Gradient Boosting (XGBoost) to address this drawback. Recurrent Neural Networks (RNNs) are utilised for the analysis of sequential behavioural data, whereas XGBoost improves predictive accuracy by using structured features. The suggested model was assessed utilising a dataset obtained from Kaggle, resulting in exceptional performance metrics: 96.8% accuracy, 97.1% precision, 95.5% recall, and a 96.3% F1-score, accompanied with an RMSE of just 0.121. The hybrid model regularly surpassed baseline models, including Decision Trees, SVM, Random Forest, and standalone LSTM, across all criteria. The research enhances predictive modelling in education and offers a scalable, computationally efficient approach for use in practical academic settings. The study has substantial significance for data-informed teacher assessment and organisational decision-making.