This project investigates the use of deep neural network models with combinatorial bandits. Combinatorial bandits are extensively used in decision making under uncertainty, where sufficient information is not available for deciding about the best decision/action. Thus the information should be learned via sophisticated methods based on exploration-exploitation methods. In this project, we combine these methods with deep learning which can endow significant flexibility and expressiveness. Deep learning models usually yield the state of art methods for representation learning and feature extraction. In this project, we will investigate the proposed method for energy-efficient navigation of electric vehicles over transport networks. In this application, the goal is to find the most suitable path over networks that are initially unknown or only partially known.