Title: Supervised and unsupervised outlier detection machine learning algorithms in wireless sensor networks

Authors: Alaa Darabseh; Praveen Khethavath

Addresses: Department of Mathematics, Engineering and Computer Science, LaGuardia Community College, CUNY Long Island City, NY, USA ' Department of Mathematics, Engineering and Computer Science, LaGuardia Community College, CUNY Long Island City, NY, USA

Abstract: Outliers influence the quality and reliability of data obtained from wireless sensor networks (WSNs) for a variety of reasons, including the effects of harsh environments and the limitations of processing and communication capacities. In this paper, we implemented and evaluated ten machine learning techniques (both supervised and unsupervised) in terms of their ability to detect outliers in WSNs. A real-world dataset was used to evaluate the results of our study. Our findings show that the best-performing unsupervised algorithms are comparable to supervised algorithms and can be used in real-world applications. This can be highly useful because unsupervised methods overcome the limitations of supervised machine learning in WSNs by not requiring historical data to build a model to detect abnormalities quickly and accurately. As a result, communication overhead will be reduced, resulting in energy savings and increased network lifetime. In addition, the results of our research show that some algorithms outperform others. This is extremely beneficial because the selection of detection algorithms is essential in improving the performance of outlier detectors.

Keywords: wireless sensor networks; WSNs; machine learning; outlier detection techniques; sensor data; classification; data quality.

DOI: 10.1504/IJBIS.2025.144869

International Journal of Business Information Systems, 2025 Vol.48 No.3, pp.332 - 352

Received: 22 Mar 2021
Accepted: 02 Jul 2021
Published online: 06 Mar 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article