Machine Learning To Detect And Prevent Sinkhole Attacks In Cluster-Based Routing Protocol Of Wsns

Abstract

The Wireless Sensor Network (WSN) Is A Collection Of Tiny Sensor Nodes That Can Collect And Process Data And Transmit It To The Base Station. Medical, Military, And Agriculture Are Some Of The Applications Of Wsns. Due To The Nature Of Wireless Sensor Networks; Nodes Are Highly Vulnerable To Many Security Threats. Among Those Threats Is The Sinkhole Attack. A Sinkhole Is A Malicious Node That Advertises The Best Path To A Base Station And Interferes With The Data To Modify And Drop Packets, It Is Also Used To Launch Other Attacks On The Network Such As Selective Forwarding And Wormhole Attacks. For This Reason, There Is A Need To Strengthen The Security Of Wireless Sensor Networks. A Machine Learning-Based Intrusion Detection System Is An Essential Security Tool To Protect Wireless Sensor Network Infrastructure And Services From These Unpredictable And Invisible Attacks. Previous Studies Of Machine Learning Have Been Proposed For Intrusion Detection In Wireless Sensor Networks And Have Achieved Reasonable Results. However, These Works Still Need To Be More Accurate And Well-Organized Against Imbalanced Data Problems In Network Traffic. In This Paper, An Efficient Machine Learning Based Intrusion Detection System For Wsns Is Proposed To Detect And Monitor The Network Activities Against Wireless Sensor Network Routing Attacks (Sinkhole, Blackhole, Flooding, And TDMA) More Efficiently. We First Describe The Challenges In Detecting Sinkhole Attacks In Wireless Sensor Networks; Followed By An Analysis Of Methods To Classify And Detect Sinkholeattacks Including The Above-Mentioned Related Network Routing Attacks. Naive Bayes, Logistic Regression, Decision Tree Random Forest, And K-Nearest Neighbors Are Used As A Classifier Under Different Conditions To Increase The Performance Of Our Model. The WSN-DS Dataset Is Used To Train And Test Our Proposed Models After Experimenting With Various Combinations Of Supervised Classifiers, We Have Found That The Detection Rate Of The Proposed Attack Detection System Using Random Forest Was Obtaining An Accuracy Of 99.52%, Precision 99.74%, Recall 99.72%, And F1-Score 99.73% Respectively. The Experimental Outcome We Got An Accuracy Of 99.27%, Precision 99.87%, Recall 99.32%, And F1-Score 99.60% Has Done After Using SMOTE Method Which Solves The Data Imbalance Problems That Happened In The Dataset. Based On The Finding, This Research Concludes That The Random Forest Classifier Is Better Than The Other Classifiers And Recommends That Network Infrastructure Managers Customize And Upgrade Their Intrusion Detection Systems (Idss) By Employing This Classifier.