Intrusion detection in wireless sensor networks using Deep Learning
No Thumbnail Available
Date
2023
Journal Title
Journal ISSN
Volume Title
Publisher
University of Oum El Bouaghi
Abstract
The humanspecies, driven by the desire for a more adequate life, has constantlystrived to advance
andcreate a modern civilization. Technology has been an unstoppable force pushing us into a brighter
future, marked by great innovations and new challenges. Wireless Sensor Networks (WSN) have
emerged as a prominent topic in ourcontemporary society, serving as a gateway to realizing the
vision of global smart citiesthrough Internet of Things (IoT) devices. These networks are finding
applications in fields as diverse as telemedicine and smart agriculture, offeringexcitingopportunities.
However, WSNs face ongoingcybersecuritythreats. Whetherit is deliberate actions by enemies
ormismanagement of the system, the security of wireless networks is of paramount importance,
presentingsignificant challenges. The limitations inherent in sensors, includinglimited memory and
powerconsumption, makesecuritymeasures a complextask. It is critical to design security solutions
thatconsidertheseconstraints, ensuring optimal network performance withoutdelays, packetloss, or
abnormalfunctionality.
This thesis proposes an intrusion detection system (IDS) engine based on deeplearning techniques,
with the aim of achievingsuperiorpacket classification results. The NSL-KDD dataset is used to
evaluate the performance of the model. The data undergoespre-processing before beingfedinto
thedeeplearning (DNN) model. Binary classification is used to distinguishbetween normal and
abnormaltrafficusing a six-layer neural network consisting of an input layer, four hiddenlayers, and
an output layer.
The resultsobtained in thisstudyshowed high precision and accuracycompared to what the
researchersachieved in theirpublishedpapers. The model achieves an accuracy of 99.65% and an F1
score of 99.67%.