Reservoir gate opening classification using multiple classifier system with ant system-based feature decomposition
Classification of reservoir gate opening (RGO) is an important task in flood management.Reservoir water level has been used to determine the number of gates to be opened when flood is imminent to prevent disaster.Predicting the number of gates to be opened is crucial to avoid any disaster. Multiple...
Saved in:
Main Authors: | , |
---|---|
Format: | Conference or Workshop Item |
Language: | English |
Published: |
2017
|
Subjects: | |
Online Access: | http://repo.uum.edu.my/22805/1/ICOCI%202017%2099-104.pdf http://repo.uum.edu.my/22805/ http://icoci.cms.net.my/PROCEEDINGS/2017/Pdf_Version_Chap02e/PID85-99-104e.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Classification of reservoir gate opening (RGO) is an important task in flood management.Reservoir water level has been used to determine the number of gates to be opened when flood is imminent to
prevent disaster.Predicting the number of gates to be opened is crucial to avoid any disaster. Multiple classifier system has been shown to provide better classification accuracy as compared to single classifier system.However, there is no guideline on the number of classifiers to be combined and no measurement was proposed to measure the compactness of the classifiers.This study proposes an ant system-based feature decomposition approach to develop a multiple classifier ensemble for classification of
RGO.Experiments have been conducted using the k-nearest neighbour, decision tree, nearest mean classifier and linear discriminant analysis as base classifier, and performance of ant system has been compared with random subspace method.Based on the results, it can be concluded that the multiple
classifier with ant system-based feature decomposition produced better classification accuracy than random subspace method. Best classification results were obtained when multiple decision tree is constructed to make predictions of RGO with an average accuracy of 89.17%. This method is expected to be useful to apply for RGO classification and future work can be done to include rainfall precipitation besides reservoir water level. |
---|