Template-Type: ReDIF-Article 1.0
Author-Name:Junaid Bakhsh,  Shakila  Parveen  Jan,  M.  Muntazir  Khan,  M.  Fawad  Mian, Farhan  Nisar, Adnan Badshah, Daud Shah, M. Nauman Khan
Author-Email:muntazirkhan131@gmail.com
Author-Workplace-Name:Institute of Computer Science and Information Technology, ICS/IT, FMCS,the University of Agriculture, Peshawar 25130, Pakistan, Department of Information Technology,Qurtuba University,Peshawar, Pakistan, Electrical Department, University of Engineering and Technology,Peshawar, Pakistan
Title:An Intelligent Intrusion Detection System Using Ensemble Learning for Ultra-Dense IoTNetworks
Abstract:Intrusion  detection  refers  to  the  process  of  observing  and  analyzing  network  or  system incidents  in  a  perpetual  manner to  identifyunauthorized  accesses,  malicious  acts,  or violationsof the rules. It plays a pivotal role in the protection of critical information, the prevention  of  security  breaches,and  the  safety,  confidentiality,  and  availability  of  company assets. Strong methods to identify and stop harmful activity are required because cybersecurity threats have grown more complex due to the quick expansion of digital infrastructure. Various researchers  have  conducted  different  research  studies  for  intrusion  detection,and  different methodologies,along with traditional as well as machine learning models,have been applied with various datasets for the proposed task. This research aims to address these challenges by developing an  efficient  and  intelligent  intrusion  detection  system  using a  stacking  ensemble learning  approach.  The  proposed  model  integrates  multiple  base  classifiers:Decision  Tree, Naïve  Bayes,  K-Nearest  Neighbor  (KNN),  and  Linear  Discriminant  Analysis  (LDA)  to capture  diverse  decision  boundaries,  with  a  Random  Forest  acting  as  the  meta-classifier  to aggregate  and  optimize  final  predictions.  The  publicly  available  UNSW-NB15  dataset  is employed in this study for intrusion detection. Python and its libraries are used for simulation purposes. After simulation, it has been achieved that the stacked model, which combines thepredictions of multiple base learners through a meta-classifier, achieved a significantly higher accuracy  of  99.93%.  While  in  comparison,  LDA  achieved  the  highest  accuracy  of  94.25%, followed closely by SVM at 93.05%, DT at 91.00%, NB at 90.55%, and KNC at 89.81%. This demonstrates  that  ensemble  learning,  particularly  stacking,  can  effectively  leverage  the strengths of individual models to greatly enhance intrusion detection performance for complex datasets.
Keywords:Intrusion, Detection, KNN, SVM, LDA, Accuracy, Confusion Matrix
Journal: International Journal of Innovations in Science and Technology
Pages:2047-2065
Volume:7
Issue:3
Year: 2025
Month:August
File-URL:https://journal.50sea.com/index.php/IJIST/article/view/1536/2230
File-Format: Application/pdf
File-URL:https://journal.50sea.com/index.php/IJIST/article/view/1536
File-Format: text/html
Handle: RePEc:abq:IJIST:v:7:y:2025:i:3:p:2047-2065