In this digital age, data is growing as faster as unimaginable. One common problem in data mining is high dimensionality which impacts the quality of training datasets and thereafter classification models. This leads to a high risk of identifying intrusions for Intrusion Detection System (IDS).The probable solution for reducing dimensionality is feature selection. Another considerable cumbersome task for constructing potent classification models from multiclass datasets is the class imbalance. This may lead to a higher error rate and less accuracy. Therefore to resolve these problems, we investigated ensemble feature selection and ensemble learning techniques for IDS. The ensemble models will decrease the hassle of selecting the wrong hypothesis and give a better approximation of the true function. In this paper Prudent Intrusion detection system (PIDS) framework, focusing on ensemble learning is given. It is a two-phase approach. Firstly, the merging of two filtering approaches is done with Ensemble Feature Selection (EFS) algorithm. The proposed EFS algorithm is implemented based on fuzzy aggregation function Height with two filtering methods: Canberra distance and city block distance. Later on, classification with Ensemble Classification (EC) algorithm is done with the unification of Support Vector Machines (SVM), Bayesian Network (BN) and K nearest neighbor (KNN). The proposed ensemble method has attained a substantial improvement in accuracy compared to single classifiers. The experiments were performed on EFS+SVM, EFS+BN, EFS+KNN and proposed framework EFS+EC.SVM recorded an accuracy rate of 81% where K-NN recorded 82.8%, Bayes network recorded 84% and our proposed EFS+EC recorded 92%. It is evidenced from the end results that this PIDS framework excels IDS and prevail the pitfalls ofSVM, Bayes network and K-NN classifiers.
展开▼
