Optimizing Education Pathways Using Hyperparameter Tuning in Neural Architecture Search (NAS)

Abstract

In general, the conventional educational recommendation models, however, cannot adequately describe the complex nonlinear learning behaviors and manual designs of neural networks are required. The goal of this work is to create an intelligent Hyperparameter-Tuned Neural Architecture Search for Education Pathway Optimization (HT-NAS-EPO) framework that combines Differentiable Architecture Search (DARTS) and Bayesian hyperparameter optimization to enhance personalized educational pathway recommendations. The proposed framework was assessed with 12,500 student data from Learning Management Systems from three academic institutions. The number of valid records after preprocessing was 11,847, each having 38 engineered features, for experimentation. Min-Max normalization and one-hot encoding were used to process the datasets, while the architecture discovery method was conducted with DARTS and hyperparameter optimization was performed with Optuna using the Bayesian Optimization technique. One-hot encoding and Min-Max normalization were used for data processing, DARTS was used for architecture discovery, and Bayesian Optimization utilizing Optuna was used for hyperparameter optimization. Several performance criteria, including accuracy, precision, recall, and F1-Score, were used to train and evaluate an optimized model. According to experimental data, the suggested framework performed better across all evaluation metrics. The optimized HT-NAS-EPO model has weighted average precision, recall, and F1-score of 93.85%, 94.10%, and 93.97%, respectively. When compared to random architectural search strategies, the total model accuracy and F1-Score were increased by 3.54% using Bayesian Optimization and by 5.18% utilizing architecture search based on DARTS. Additionally, the model had great generalization capacity and did not overfit the training data, as evidenced by the convergence of the final training loss of 0.0834 and validation loss of 0.0971. It has been shown that neural architecture search combined with intelligent hyperparameter tuning is an effective and scalable method for learning pathway optimization in adaptive educational recommendation systems, leading to a more precise, customized, and data-driven learning pathway optimization.