1. Title, Journal and Authors

Title: Online classification of contaminants based on Multi-Classification Support Vector Machine using conventional water quality sensors

Journal: Sensors

Authors: Pingjie Huang, Yu Jin, Dibo Hou*, Jie Yu, Dezhan Tu, Yitong Cao and Guangxin Zhang

State Key Laboratory of Industrial Control Technology, College of Control Science and Engineering, Zhejiang University

 

2. Summary

  In this paper, the multi-classification support vector machine was used to obtain the possibility for contaminants belonging to a category. The cosine distance classification was used for comparison with it.

 

  Off-line training was progressed and then, on-line classification was done to obtain the precise classification.

 

  In SVM, support vectors are used to determine the classification decision boundary. Its objective is to find the maximum marginal hyperplane (MMH). In this paper, a sigmoid function was employed as a kernel function. And, original binary SVM was extended to the multi-classification model. Parameters used to calculate the effectiveness of SVM were kernel function γ and the penalty parameter C. Accuracy was calculated with five-fold cross-validation. Then, classification performance was calculated with a confusion matrix.

 

  A total of 784 samples were taken. Among them, 610 samples were used to training and 174 samples were used to testing.

There were evident differences between SVM and cosine distance classification outside the range of pollutant library. So, the on-line classification method could effectively reduce the influences of concentration.

 

  Almost support vectors were obtained at low concentrations. In this result, influences of concentration were decreased and the precision of classification was increased.

 

  Finally, after analyzing the possibility outputs, it was obtained that class Ⅰ was 167 in the total 174 test samples. Class Ⅰ showed the high maximum probability and deviation. It meant it was similar to the real concentration value and others lower than it showed indistinct classification.

 

3. Application to research

We can utilize extended multi-classification SVM model for our research to classify many components.

 

4. Contact

Dae Seong Jeong / Intern student

Earth Science and Environmental Engineering Concentration

Gwangju Institute of Science and Technology (GIST)

123 Cheomdangwagi-ro, Buk-gu, Gwangju, 61005, Republic of Korea 

 

Phone : +82-10-2003-7860

E-mail : jeongds92@gist.ac.kr