Laporan Tahunan UPNM 2020

191 Un i v e r s i t i P e r t a h a n a n Na s i o n a l Ma l a y s i a L A P O R A N T A H U N A N 2 0 2 0 Penerbitan Ilmiah Judul: Swarm Intelligence in Data Classification e-ISBN: 978-967-2414-09-4 Format/Bahasa: epub/Bahasa InggerisPengarang: Mohd `Afizi Mohd Shukran, Mohd Sidek Fadhil Mohd Yunus, Muhammad Naim Abdullah Ringkasan: Data classification involves solving problems by analysing data already present in databases. Due to the explosive growth of both business and scientific databases, extracting efficient classification rules from such databases has become an important task. This is because classification technique is an important form of knowledge extraction and can help to make key decisions. Nevertheless, classification technique can be improved by integrating the latest technology, namely, Swarm Intelligence. This study proposes two types of classification techniques: Artificial Bee Colony, and Intelligent Dynamic Swarm, which are both based on Swarm Intelligence. This is because Swarm Intelligence has the capability to adapt well in changing environments and is immensely flexible and robust. The first swarm based classifier involves using the advantages of Artificial Bee Colony as an optimization tool to do the data classification. This proposed Artificial Bee Colony based classifier has been implemented to the Anomaly based Network Intrusion Detection System. To our knowledge, it is the first time that the Artificial Bee Colony technique has been applied to solve the network intrusion detection problem. Another swarm based classifier that has been proposed in this study is a novel Intelligent Dynamic Swarm, which is based on Particle Swarm Optimization. Unlike a conventional Particle Swarm Optimization algorithm, this novel algorithm can directly cope with discrete variables. In addition, Intelligent Dynamic Swarm can successfully avoid premature convergence, which is considered a serious drawback of traditional Particle Swarm Optimization. These two proposed new swarm based data classification algorithms have been evaluated using the UCI data set, KDD-99 datasets developed by MIT Lincoln Labs, and the pre-processed image data. The experimental results showed that both the Anomaly based Network Intrusion Detection System and Intelligent Dynamic Swarm are robust and able to achieve high classification accuracy in a changing environment within the data instances. Therefore, both proposed classifiers can provide a promising direction for solving complex problems that may not be solved by traditional approaches. Layout (Annual Report UPNM 2020).indd 191 07/03/2022 11:50 AM