The growing need to make accurate energy consumption forecasting has prompted improvements in machine learning. The study examines the suitability of both conventional ANN and enhanced version, which is DNN, in predicting client’s electricity usage patterns. We investigate the architectures, training schemes, and the final performances of each of these methods using a large dataset that covers different energy consumption scenarios. The study starts with a comprehensive review of traditional and modern neural network models. The models are ANNs and DNN, and the characteristic features of their train and test phases are shown in this section. Complexity of pattern recognition in deep learning models, accompanied with relevant case-studies that highlight application of these approaches across different environments. As we study throughout, the issues of interpretability and scalability with regards to neural network models are addressed and insights offered. Furthermore, this research also focuses on cyber security and utilizes the anomaly detection strategies. Relating anomaly detection in energy use to cyber security, we show how important these approaches are in preventing attacks. We bring cybersecurity considerations towards the debate around the role of deep learning for real-world energy sector’s challenges. For instance, this applies to artificial and deep neural networks for forecasting consumption of energy and improving cyber-security measures. In looking forward, our finding will assist in refinements and improvements of the models and provide new methods to increase accuracy in modelling with a particular focus on big data centres and their cyber-security.