A persistent issue in classifying data traffic has been created by repetitive and pointless data characteristics. When dealing with enormous data, these characteristics not only make identification slower, but they also make it more difficult for a classifier to give precise judgements. In this research, we provide an algorithm based on mutual knowledge that chooses the best feature for classifying by analytical means. The feature selection approach, which is based on similar information, can tackle elements with both linear and nonlinear dependencies in the data. In scenarios involving network intrusion detection, its efficacy is assessed. We develop an intrusion detection system (IDS) called Least Square Support Vector Machine based IDS, which is based on the extracted features using our recommended feature selection approach (LSSVM-IDS). Datasets like KDD Cup 99, NSL-KDD that help in analyzing the performance of LSSVM IDS. The assessment findings demonstrate that, in comparison to the province technique, our feature selection algorithm provides more crucial characteristics that help LSSVM-IDS improves performance and reduces the complexity.