Schwabacher [7] addressed pruning for distance-based outlier algorithms applied for
large real-world data sets introduced.
Another clustering philosophy is based on density of points. Examples of such a
philosophy are given by Breunig et al. [8, 9], in relation to relative density. Jin et al.
[10] showed how some of the calculations of the density function can be skipped or
estimated based on distribution type. Tang et al. [11] further improved clustering by
adding the idea of a connectivity-based outlier factor, which refers to the number of
connections between points.
With regard to WQE specifically, a major work has been done under the EPA
(US Environment Protection Agency) framework and including the development of the
Canary software. Canary is a free software tool developed by the EPA which aims to
detect abnormality in water network [12, 13]. An extensive comparison work (https://
www.epa.gov/sites/production/files/2015-07/documents/water_quality_event_detec
tion_system_challenge_methodology_and_findings.pdf) done by the EPA compared
the Canary algorithm to other tools. It was reported based on comparison between
actual and expected value of water quality measurements that this algorithm does not
have general superiority over other machine learning algorithms. That is, using linear
regression to predict water quality does not do better than other methods. In this specific
report, the Canary was ranked third in many tests based on the amount of false alarm
reported. Another conclusion which was pointed by this report is the fact that the
Canary has a narrow time window from which its algorithm learns and hence can’t
provide conclusions based on large data set. Abnormality detection has two main
methods as was detailed by Clark and Hakim [14]: supervised and unsupervised.
Examples for WQE detection have also been provided by Skadsen [15], Story et al.
[16], Yang et al. [17], Chang et al. [18], and Helbling and VanBriesen [19]. A set of
more commercial reports with many useful pieces of information can be found at the
following EPA website: http://www.epa.gov/nhsrc/pubs.html.
The current chapter examines the efficiency of radial basis function (RBF) as an
identification and classification tool for WQE. RBF is known in the literature as a
tool for abnormality detection in other areas. Examples are given by Mellisa et al.
[20] for mammograms classification; Padmapriya et al. [21] for brain tumor detection; Rajab and Salleh [22] for classification of diabetes; Mansourkhaki et al. [23] for
traffic prediction; and Chun-Cheng Lin and Weichih Hu [24] for detection of
abnormal intra-QRS pulses.
The current chapter demonstrates the implementation of RBF in the WQE
detection domain. The structure of the chapter is as follows: the second section of
the chapter gives a description of RBF and RBF networks. The third section gives a
description of parameter values selection. Section 4 describes a lab-based example.
Section 5 is the analysis of the data set with the traditional form of RBF. Section 6
shows a similar analysis with several other forms of RBF kernel function. Section 7
concludes the chapter.
Using Radial Basis Function for Water Quality Events Detection
143
large real-world data sets introduced.
Another clustering philosophy is based on density of points. Examples of such a
philosophy are given by Breunig et al. [8, 9], in relation to relative density. Jin et al.
[10] showed how some of the calculations of the density function can be skipped or
estimated based on distribution type. Tang et al. [11] further improved clustering by
adding the idea of a connectivity-based outlier factor, which refers to the number of
connections between points.
With regard to WQE specifically, a major work has been done under the EPA
(US Environment Protection Agency) framework and including the development of the
Canary software. Canary is a free software tool developed by the EPA which aims to
detect abnormality in water network [12, 13]. An extensive comparison work (https://
www.epa.gov/sites/production/files/2015-07/documents/water_quality_event_detec
tion_system_challenge_methodology_and_findings.pdf) done by the EPA compared
the Canary algorithm to other tools. It was reported based on comparison between
actual and expected value of water quality measurements that this algorithm does not
have general superiority over other machine learning algorithms. That is, using linear
regression to predict water quality does not do better than other methods. In this specific
report, the Canary was ranked third in many tests based on the amount of false alarm
reported. Another conclusion which was pointed by this report is the fact that the
Canary has a narrow time window from which its algorithm learns and hence can’t
provide conclusions based on large data set. Abnormality detection has two main
methods as was detailed by Clark and Hakim [14]: supervised and unsupervised.
Examples for WQE detection have also been provided by Skadsen [15], Story et al.
[16], Yang et al. [17], Chang et al. [18], and Helbling and VanBriesen [19]. A set of
more commercial reports with many useful pieces of information can be found at the
following EPA website: http://www.epa.gov/nhsrc/pubs.html.
The current chapter examines the efficiency of radial basis function (RBF) as an
identification and classification tool for WQE. RBF is known in the literature as a
tool for abnormality detection in other areas. Examples are given by Mellisa et al.
[20] for mammograms classification; Padmapriya et al. [21] for brain tumor detection; Rajab and Salleh [22] for classification of diabetes; Mansourkhaki et al. [23] for
traffic prediction; and Chun-Cheng Lin and Weichih Hu [24] for detection of
abnormal intra-QRS pulses.
The current chapter demonstrates the implementation of RBF in the WQE
detection domain. The structure of the chapter is as follows: the second section of
the chapter gives a description of RBF and RBF networks. The third section gives a
description of parameter values selection. Section 4 describes a lab-based example.
Section 5 is the analysis of the data set with the traditional form of RBF. Section 6
shows a similar analysis with several other forms of RBF kernel function. Section 7
concludes the chapter.
Using Radial Basis Function for Water Quality Events Detection
143