Detail publikace

Convergence Optimization of Backpropagation Artificial Neural Network Used for Dichotomous Classification of Intrusion Detection Dataset

Originální název

Convergence Optimization of Backpropagation Artificial Neural Network Used for Dichotomous Classification of Intrusion Detection Dataset

Anglický název

Convergence Optimization of Backpropagation Artificial Neural Network Used for Dichotomous Classification of Intrusion Detection Dataset

Jazyk

en

Originální abstrakt

There are distinguished two categories of intrusion detection approaches utilizing machine learning according to type of input data. The first one represents network intrusion detection techniques which consider only data captured in network traffic. The second one represents general intrusion detection techniques which intake all possible data sources including host-based features as well as network-based ones. The paper demonstrates various convergence optimization experiments of a backpropagation artificial neural network using well know NSL-KDD 1999 dataset, and thus, representing the general intrusion detection. Experiments evaluating usefulness of stratified sampling on input dataset and simulated annealing employed into the backpropagation learning algorithm are performed. Both techniques provide improvement of backpropagation's learning convergence as well as classification accuracy. After 50 training cycles, classification accuracy of 84.20% is achieved when utilizing stratified sampling and accuracy of 86.5% when both stratified sampling and simulated annealing are used. In contrast, the backpropagation by itself reaches only 76.63% accuracy. Comparing to state-of-the-art work introducing the NSL-KDD dataset, there is achieved accuracy higher about 4.5%.

Anglický abstrakt

There are distinguished two categories of intrusion detection approaches utilizing machine learning according to type of input data. The first one represents network intrusion detection techniques which consider only data captured in network traffic. The second one represents general intrusion detection techniques which intake all possible data sources including host-based features as well as network-based ones. The paper demonstrates various convergence optimization experiments of a backpropagation artificial neural network using well know NSL-KDD 1999 dataset, and thus, representing the general intrusion detection. Experiments evaluating usefulness of stratified sampling on input dataset and simulated annealing employed into the backpropagation learning algorithm are performed. Both techniques provide improvement of backpropagation's learning convergence as well as classification accuracy. After 50 training cycles, classification accuracy of 84.20% is achieved when utilizing stratified sampling and accuracy of 86.5% when both stratified sampling and simulated annealing are used. In contrast, the backpropagation by itself reaches only 76.63% accuracy. Comparing to state-of-the-art work introducing the NSL-KDD dataset, there is achieved accuracy higher about 4.5%.

BibTex


@article{BUT133490,
  author="Ivan {Homoliak} and Dominik {Breitenbacher} and Petr {Hanáček}",
  title="Convergence Optimization of Backpropagation Artificial Neural Network Used for Dichotomous Classification of Intrusion Detection Dataset",
  annote="There are distinguished two categories of intrusion detection approaches
utilizing machine learning according to type of input data. The first one
represents network intrusion detection techniques which consider only data
captured in network traffic. The second one represents general intrusion
detection techniques which intake all possible data sources including host-based
features as well as network-based ones. The paper demonstrates various
convergence optimization experiments of a
backpropagation artificial neural network using well know NSL-KDD 1999 dataset,
and thus, representing the general intrusion detection. Experiments evaluating
usefulness of stratified sampling on input dataset and simulated annealing
employed into the backpropagation learning algorithm are performed. Both
techniques provide improvement of backpropagation's learning convergence as well
as classification accuracy. After 50 training cycles, classification accuracy of
84.20% is achieved when utilizing stratified sampling and accuracy of 86.5% when
both stratified sampling and simulated annealing are used. In contrast, the
backpropagation by itself reaches only 76.63% accuracy. Comparing to
state-of-the-art work introducing the NSL-KDD dataset, there is achieved accuracy
higher about 4.5%.",
  address="NEUVEDEN",
  chapter="133490",
  doi="10.17706/jcp.12.2.143-155",
  edition="NEUVEDEN",
  howpublished="online",
  institution="NEUVEDEN",
  number="2",
  volume="12",
  year="2017",
  month="march",
  pages="143--155",
  publisher="NEUVEDEN",
  type="journal article in Web of Science"
}