Intrusion Detection System Against Malign Packets—A Comparative Study Between Autoencoder and Ensemble Model

An intrusion detection system generally involves the use of a traditional cryptography algorithm. However, with rise in technology, number of attackers seeking privacy increases. Therefore, it is necessary to have a robust intrusion detection system which can prevent future attacks which are alien to the system. Although traditional approaches have achieved great success towards eliminating adversarial contamination, our robust autoencoder is capable to eliminate unknown attacks with greater success. In addition, we have compared our model with respect to random forest algorithm. Our experimental research shows variance in accuracy of both the algorithms by 7–8%. The key factor that directly affects the accuracy of the model is the threshold value which was determined using stochastic approaches. Thus, the obtained results concerning the detection ability of the autoencoder, compared to that of random forest algorithm, is also encouraging, with autoencoder limiting the percentage of false positive attacks which is proved to be fatal for an intrusion detection system.