Simulation Analysis of Optical Sensor-Based Intrusion Detection Using Machine Learning Algorithms

EasyChair Preprint 10122

Abstract

The maintenance of security in varied situations depends on intrusion detection. In this study, we assess three machine-learning systems' abilities to identify intrusions using data from optical sensors. The algorithms tested are Ridge Classifier, k-Nearest Neighbor (KNN), and a neural network. The system uses data collected from Optical Time Domain Reflectometer (OTDR) machines, which receive data from optical fiber sensors laid on the ground or walls/fences. The difference in the amplitude between the OTDR data traces that result from an intruder’s movement disrupting the optical fiber signals is utilized to identify intrusions. The system preprocesses the data, and the three machine learning models are trained on the preprocessed data. Our study shows that ANN outperforms Ridge Classifier and the ANN in terms of accuracy, achieving 93% accuracy compared to Ridge Classifier’s 92.5% and the neural network's 91%. These results indicate that KNN is a promising algorithm for intrusion detection using optical sensors.

Keyphrases: ANN, Intrusion Detection, KNN, Ridge Classifier, optical sensors

@booklet{EasyChair:10122,
  author    = {Kavitha Thandapani and Akshit Kasanagottu and Jayasurya Pasupula and Sriharsha Daggubati},
  title     = {Simulation Analysis of Optical Sensor-Based Intrusion Detection Using Machine Learning Algorithms},
  howpublished = {EasyChair Preprint 10122},
  year      = {EasyChair, 2023}}