Enhancement of Intrusion Detection Dataset in Wireless Sensor Network using RNS - Feature Conversion with Stack Ensemble Technique


  • I.R. Idowu 1Federal College of Animal Health &Production Technology, Moor Plantation, Ibadan, Nigeria
  • A.W. Asaju-Gbolagade, University of Ilorin, Ilorin, Nigeria. 3Kwara State University Ilorin, Nigeria
  • K.A. Gbolagade Kwara State University Ilorin, Nigeria


Ensemble Classifiers, Intrusion Detection System, Particle Swarm Optimization, Residue Number System, Wireless Sensor Network


This research presents a feature selection and conversion technique for Wireless sensor network (WSN) for the enhancement of classification and detection of intrusions. There are many different approaches and datasets, but the performance of the current Intrusion detection systems (IDSs) does not seem to be sufficient because there are so many data volumes that need to be processed in a less ample time that it is beyond the capacity of the most widely used hardware and software tools. However, computational challenges and inadequate quality still exist in state-of-art of feature selection approaches in IDS. In order to effectively and optimally minimize the feature size of the data dimensions, the Particle Swamp Optimization (PSO) approach was presented, thereafter Residue Number System (RNS) was used to further convert the selected features from the dataset using moduli of {2(n+1) - 1, 2(n) - 1, 2(n)} to residues in order to reduce large weighted number to several small numbers and
enhance the power consumption and improve the time complexity further. The outcomes demonstrate that Case 1; a composite of Z-Score, PSO, RNS, and Ensemble Classifier performed better than the case without the procedure of features conversion in Case2;(Z-Score + PSO+ Ensemble Classifier),in terms of the well-known UNSW-NB 15 dataset's classification accuracy, error rate, sensitivity, specificity, and training time. The classification accuracy shows the highest classification rate for CASE 1 to be 97.4736% and 95.3602% for CASE 2. The result shows a clear cut difference of over 2% in variation indicating the prominence of feature
conversion in WSN dataset