Nowadays, systems and networks play a key role in human lives. This role has made any threat and cyber-attack more effective and destructive. Along with the rapid growth of networks and computer systems, threat factors have improved the techniques and approaches used to perform their attacks. On the other hand, the conventional approaches of abnormal based detection systems have been widely studied based on machine learning techniques over the past 40 years. The continual process, high volume, applying modern methods in performing attacks and improving hardware and software data analysis capabilities have led machine learning approaches to overcome the issues of intrusion detection. In a machine learning area, the deep learning approaches could exhibit high efficiency and be successful in solving those problems which were the aim of active researches in the recent decades. The major goal of present research is to apply deep learning to solve the existing issues in abnormality based intrusion detection area and present a highly efficient intrusion detection system. Also, it is shown that deep learning is an efficient method in the field of anomaly based intrusion detection and is able to improve efficiency over other existing methods. The experimental evaluation of the proposed method has been done on UNSW- NB15 dataset and the 99% detection rate and 0.6% false alert rate have been obtained. Keywords: Anomaly based intrusion detection system, Deep learning, Cyber attack, UNSW-NB15