Improved departure from nucleate boiling prediction in rod bundles using a physics-informed machine learning-aided framework