Log Based Intrusion Detection System Using Hybrid Machine Learning Approach
Cloud computing is getting popularity whereas evolving security threats and performance are the two important main constraints faced by different scale of cloud providers and clients. Now a day large scale internet user evolution proportionally elevated the computer security care. An intrusion detection system (IDS) is widely used approach capable of distinguishing between attacks and normal network connections signature. The designed intrusion detection system model is trained and evaluated using KDD datasets, which consists different network log features. Researchers are focusing on hybrid approaches to model IDS as it can combine the advantages of two or more algorithms. We applied mathematical, statistical techniques and machine learning approach to increase the efficiency of intrusion detection rate. Feature Discretizing, Scaling, and Covariance matrix are used as data pre-processing stage for principal component analysis (PCA). PCA is introduced with the motive to extract a low-dimensional set of features from a high dimensional dataset. Eigenvector and eigenvalue used to find the significant principal components. The original KDD dataset contains 42 dimensions in which we extracted 20 principal components using PCA. Based on eigenvector result the selected principal components account 95.35% of information which shows insignificant information loose. The reduced dimension improves the performance of our training model. The evaluation of our model performance is demonstrated on the KDD’99 dataset. From the experiment, we get improved result with a promising performance in relative to existing works on IDS. After advanced modeling with different machine learning algorithms and extensive experiment we propose the model with moderate detection capability and reduced false alarm rate.