Ad Hoc Networks 49 (2016) 70–89 Contents lists available at ScienceDirect Ad Hoc Networks journal homepage: www.elsevier.com/locate/adhoc Feature selection for performance characterization in multi-hop wireless sensor networks Athanasia Panousopoulou a,∗ , Mikel Azkune b , Panagiotis Tsakalides a,c a Institute of Computer Science, Foundation of Research and Technology, Hellas, Heraklion, Crete, Greece b Applied Photonics Group,Faculty of Engineering, University of the Basque Country, Bizkaia, Spain c Department of Computer Science, University of Crete, Heraklion, Crete, Greece a r t i c l e i n f o Article history: Received 8 October 2015 Revised 28 April 2016 Accepted 20 June 2016 Available online 21 June 2016 Keywords: Wireless sensor networks Unsupervised learning Network measurement and analysis Testbeds and experimental evaluation a b s t r a c t Current trends in Wireless Sensor Networks are faced with the challenge of shifting from testbeds in con- trolled environments to real-life deployments, characterized by unattended and long-term operation. The network performance in such settings depends on various factors, ranging from the operational space, the behavior of the protocol stack, the intra-network dynamics, and the status of each individual node. As such, characterizing the network’s high-level performance based exclusively on link-quality estima- tion, can yield episodic snapshots on the performance of specific, point-to-point links. The objective of this work is to provide an integrated framework for the unsupervised selection of the dominant features that have crucial impact on the performance of end-to-end links, established over a multi-hop topology. Our focus is on compressing the original feature vector of network parameters, by eliminating redundant network attributes with predictable behavior. The proposed approach is implemented alongside different cases of protocol stacks and evaluated on data collected from real-life deployments in rural and industrial environments. Discussions on the efficacy of the proposed scheme, and the dominant network character- istics per deployment are offered. © 2016 Elsevier B.V. All rights reserved. 1. Introduction Over the past years, Wireless Sensor Networks (WSN) have been closing the gap between theory and application in real-life scenarios, thereby gaining prominence as the key enabling tech- nology for addressing significant societal challenges [1–3]. Exploit- ing WSN-based schemes for solving modern engineering problems, intensifies the necessity of transiting from episodic sampling to truly pervasive paradigms relying on resilient, long-term, and unat- tended operation. As such, monitoring and characterizing the per- formance of the network in realistic deployments is gaining in- creasing interest [4], as a process influenced by multiple factors. Recent works [5,6] emphasize the necessity for providing system- atic tools, capable of capturing a variety of different aspects of ra- dio transmission and wireless network deployments. High level re- quirements, such as application-driven positioning and scale, can impact the network performance [7]. The behavior of multi-hop links is dominated by the dynamics of wireless connectivity and power autonomy, even when the sensor nodes are in fixed posi- tions [8]. The combination of the operational space and the hard- ∗ Corresponding author. E-mail address: apanouso@ics.forth.gr (A. Panousopoulou). ware characteristics become key factors. Finally, from the perspec- tive of application-driven deployments, guaranteeing the desired Quality of Service is considered more important than the low-level details of sophisticated protocol stacks. Addressing the aforementioned challenges can be acceler- ated by employing passive monitoring mechanisms that observe the performance of user-designated end-to-end links. By the term “end-to-end”, we refer to network links, which are built over a multi-hop network topology and are responsible for the application-driven data flow. Opposed to the well-studied point- to-point links that are formulated at the Physical layer, and are ca- pable of link quality estimation between 1-hop neighbors, end-to- end links expand towards two different directions: (a) across dif- ferent sides of the network, exceeding the constrained limits of 1- hop neighborhoods, (b) across different layers of a fully functional protocol stack, ranging from the Physical to the Transport and Ap- plication layers. As such, end-to-end links convey a larger volume of information than the one captured by point-to-point, low-level links. Thus, the systematic study of their performance could pro- vide the means for understanding the multi-dimensional behavior of the entire network. Enabling the systematic collection and process of sufficient amounts of data for characterizing the performance of end-to- http://dx.doi.org/10.1016/j.adhoc.2016.06.011 1570-8705/© 2016 Elsevier B.V. All rights reserved.