Outlier detection in wireless sensor network based on time series approach
Sensory data inWireless Sensor Network (WSN) is not always reliable because of open environmental factors such as noise, weak received signal strength or intrusion attacks. The process of detecting highly noisy data and noisy sensor node is called outlier detection. Outlier detection is one of the f...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | http://eprints.utm.my/id/eprint/98189/1/MahmoodSafaeiPSC2019.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Sensory data inWireless Sensor Network (WSN) is not always reliable because of open environmental factors such as noise, weak received signal strength or intrusion attacks. The process of detecting highly noisy data and noisy sensor node is called outlier detection. Outlier detection is one of the fundamental tasks of time series analysis that relates to predictive modeling, cluster analysis and association analysis. It has been widely researched in various disciplines besides WSN. The challenge of noise detection in WSN is when it has to be done inside a sensor with limited computational and communication capabilities. Furthermore, there are only a few outlier detection techniques in WSNs and there are no algorithms to detect outliers on real data with high level of accuracy locally and select the most effective neighbors for collaborative detection globally. Hence, this research designed a local and global time series outlier detection in WSN. The Local Outlier Detection Algorithm (LODA) as a decentralized noise detection algorithm runs on each sensor node by identifying intrinsic features, determining the memory size of data histogram to accomplish effective available memory, and making classification for predicting outlier data was developed. Next, the Global Outlier Detection Algorithm (GODA)was developed using adaptive Gray Coding and Entropy techniques for best neighbor selection for spatial correlation amongst sensor nodes. Beside GODA also adopts Adaptive Random Forest algorithm for best results. Finally, this research developed a Compromised SensorNode Detection Algorithm (CSDA) as a centralized algorithm processed at the base station for detecting compromised sensor nodes regardless of specific cause of the anomalies. To measure the effectiveness and accuracy of these algorithms, a comprehensive scenario was simulated. Noisy data were injected into the data randomly and the sensor nodes. The results showed that LODA achieved 89% accuracy in the prediction of the outliers, GODA detected anomalies up to 99% accurately and CSDA identified accurately up to 80% of the sensor nodes that have been compromised. In conclusion, the proposed algorithms have proven the anomaly detection locally and globally, and compromised sensor node detection in WSN. |
---|