A significant field of investigation in network security is intrusion detection. The detection of abnormalities in network data is a typical method for intrusion detection, but network threats are developing at an extraordinary speed. The system is exposed to attacks due to the discrepancy between both the development of threats and the network's existing detection reaction time. A variety of machine learning methods have been created over time to identify network breaches using packet forecasting. These methods rely on techniques that can learn automatically from data without explicitly programmed. This is especially practical given the diverse nature of the traffic. Nevertheless, despite these benefits, abuse detection still outperforms anomaly detection systems in the actual world. The principal cause of the low adoption of anomaly-based intrusion detection system is the issue of the significant false positive rate. On a network with considerable traffic, even a 1% false positive rate can result in so many false alerts that an administrator is unable to handle them. We offer suggestions for applying deep learning machine learning method to increase the accuracy of anomaly-based IDS detection for implementation on actual networks. In our research methodology we proposed intrusion detection system in two phases using various datasets namely KDDcup99, botnet, ISCX, WSNtrace, NSLKDD, NUSW-NB15 and real time twitter data. In phase 1, we proposed Intrusion detection system using Genetic Algorithm and various Machine learning techniques such as J48, Artificial Neural Network and Random Forest. In this phase, we have performed various experiments for evaluating performance of our proposed system by considering various parameters like using different population size, threshold, datasets etc. In the phase 2, we have evaluated proposed RNN-LSTM for various functions like sigmoid, tanh and ReLu using different cross validation. Our proposed system can generate its own rules. It can detect DOS, Root to login, probe, User to Root, network attack, passive attack, active attacks as well as unknown attacks. It is observed from the experimental findings that detection rate of denial of service attack is 96.9 % which is high as compared to other attacks and the accuracy rate of RNN-LSTM (ReLU) using 20-fold cross validation is 97.95 % which is high as compared to RNN-LSTM (Sigmoid) and RNN-LSTM (Tanh) when different cross validations like 10-fold, 15-fold and 20-fold are used.

Intrusion detection system, machine learning, network attacks, DDoS, Unknown attacks, feature extraction, classification, deep learning, network security