The popularity of mobile ad hoc networks (MANETs) has grown significantly due to their cheap and ease of deployment. However, they appear to be more vulnerable to various attacks due to the open access medium and the dynamically changing network topology such as denial of service, packet dropping, routing disruption and resource consumption. Intrusion detection system plays an important role to safeguard security in MANETs to monitor network traffic and make a proper analysis of the network. There are various research studies to prevent and detect such types of attacks, however still needs additional studies because of the continuous expansion of network data, a large amount of non-linear network data brings new challenges to intrusion detection. This study aims to preprocess and select the important features to enhance the performance of detection model. We used the Bidirectional long-short term memory (BiLSTM) deep learning algorithm applied on NSL-KDD dataset we have achieved the highest prediction accuracy rate 99.673% and lower false positive rate of 0.0118%.