An Effective Comparative Analysis of Data Preprocessing Techniques in Network Intrusion Detection System Using Deep Neural Networks

Recently machine learning algorithms are utilized for identifying network threats. Threats otherwise called as intrusions, will harm the network in a stern manner, thus it must be dealt cautiously. In the proposed research work, a deep learning model has been applied to recognize and categorize unanticipated and unpredictable cyber-attacks. The UNSW NB-15 dataset has a vital number of features which will be learned by the hidden layers present in the suggested model and classified by the output layer. The suitable quantity of layers, neurons in each layer and the optimizer utilized in the proposed work are obtained through a sequence of trial and error experiments. The concluding model acquired can be utilized for estimating future malicious attacks. There are several data preprocessing techniques available at our disposal. We used two types of techniques in our experiment: 1) Log transformation, MinMaxScaling and factorize technique; and 2) Z-score encoding and dummy encoding technique. In general, the selection of data preprocessing techniques has a direct impact on the output performed by any machine learning process and our research, attempts to prove this concept.