Q346 : An Efficient Machine Learning-baxsed Anomaly Detection Method in Wireless Sensor Network
Thesis > Central Library of Shahrood University > Computer Engineering > MSc > 2026
Authors:
[Author], Alireza Tajary[Supervisor], Mansoor Fateh[Advisor], Elias Khaje Karimi[Advisor]
Abstarct: Abstract Wireless Sensor Networks (WSNs) have become an essential technology in numerous applications, including environmental monitoring, industrial automation, healthcare, and smart cities. However, their deployment in dynamic and resource-constrained environments makes them vulnerable to abnormal sensor behavior, communication failures, and unexpected environmental disturbances. Therefore, accurate anomaly detection has become a critical requirement for improving network reliability, maintaining data integrity, and supporting intelligent decision-making. This research proposes an effective hybrid anomaly detection frxamework that integrates deep learning and machine learning techniques for Wireless Sensor Networks. The proposed methodology employs a CNN-Attention Autoencoder to automatically extract high-level nonlinear feature representations from sensor measurements. Principal Component Analysis (PCA) is subsequently applied to reduce feature dimensionality while preserving the most informative characteristics. Finally, XGBoost is utilized as the classification model to distinguish normal observations from anomalous events. The proposed frxamework was evaluated using the Intel Berkeley Research Lab dataset, containing approximately 2.3 million environmental sensor readings collected from 54 wireless sensor nodes under realistic operating conditions. Comprehensive preprocessing, feature engineering, and statistical labeling procedures were performed prior to model training. The proposed frxamework was compared with several conventional anomaly detection approaches, including Isolation Forest, Local Outlier Factor, Random Forest, and Extra Trees. Experimental results demonstrated that the hybrid frxamework consistently achieved superior classification performance, with high Accuracy, Precision, Recall, F1-Score, ROC-AUC, and Precision–Recall characteristics. The findings confirm that combining deep feature extraction with dimensionality reduction and gradient boosting significantly enhances anomaly detection capability while reducing computational complexity. Consequently, the proposed frxamework provides a robust, scalable, and efficient solution suitable for practical Wireless Sensor Network applications.
Keywords:
#_Wireless Sensor Networks #Anomaly Detection #Deep Learning #CNN-Attention Autoencoder #Principal Component Analysis #XGBoost #Machine Learning #Feature Extraction_ Keeping place: Central Library of Shahrood University
Visitor: