Ensemble-based methods for intrusion detection

摘要: Abstract The master thesis focuses on ensemble approaches applied to intrusion detection systems (IDSs). approach is a relatively new trend in artificial intelligence which several machine learning algorithms are combined. main idea exploit the strengths of each algorithm obtain robust classifier. Moreover, ensembles particularly useful when problem can be segmented into subproblems. In this case, module ensemble, include one or more algorithms, assigned particular subproblem. Network attacks divided four classes: denial service, user root, remote local and probe. One designed work itself an decision trees specialized class attacks. inner structure uses bagging techniques increase accuracy IDS. Experiments showed that IDSs better results treated as separate handled by algorithms. This have also concluded these need trained with specific subsets fea- tures selected according their relevance attack being detected. efficiency highlighted. all experiments, was able bring down number false positives negatives. However, we observed limitations KDD99 dataset. particular, distribution examples between training set test made difficult evaluation for attack.