Innovative IDS Model Incorporating Machine Learning and Feature Reduction Mechanisms

It is becoming more difficult to promptly warn any harmful activity to avert the loss of sensitive data and money. This is because the quantity of new assaults on ever-growing system circulation is rising. Consequently, when it comes to network security, ID is one of the most important areas of apprehension. The real-world techniques that are being practiced frequently nowadays are mostly based on the technique known as Anomaly-based Network ID. The effectiveness of the strategies is therefore relative to the performance of the dataset that was used off to test the strategies. Unfortunately, this dataset does not, in most cases, paint the real picture of traffic flow in the network. It is for this reason that this research is about the study of several ML algorithms that is used in IDS. These algorithms are assessed on two datasets like the current network traffic in the real world and a modified KDD’99 dataset called NSL-KDD. This study was also planned to design a new composite model of IDS if explorations of various intruder detection systems on both sets of data showed promising results. Thus, this creative integration, employing stacking technique, of decision tree and random forest, yields an accuracy of 85 per cent. The recall is 2% while the precision is 86. G Wav: 1% and NSL-KDD: 2%. Also, it provides an accuracy of 98.99% and a precision of 98.99% on the CICIDS2017 dataset while obtaining such performances.