Mohamad Mohsin, Mohamad Farhan and Hamdan, Abdul Razak and Abu Bakar, Azuraliza (2014) An evaluation of feature selection technique for dendrite cell algorithm. In: International Conference on IT Convergence and Security (ICITCS) 2014, 28-30 Oct. 2014, Beijing.

PDF Restricted to Registered users only Download (244kB) | Request a copy

Abstract

Dendrite cell algorithm needs appropriates feature to represents its specific input signals. Although there are many feature selection algorithms have been used in identifying appropriate features for dendrite cell signals, there are algorithms that never been investigated and limited work to compare performance among them. In this study, six feature selection algorithms namely Information Gain, Gain Ratio, Symmetrical Uncertainties, Chi Square, Support Vector Machine, and Rough Set with Genetic Algorithm Reduct are examined and their effectiveness to represent dendrite cell signal are evaluated. Eight universal datasets are chosen and assessing their performance according to sensitivity, specificity, and accuracy. From the experiment, the Rough Set Genetic Algorithm reduct is found to be the most effect feature selection for dendrite cell algorithm when it generates a consistent result for all evaluation metrics. In single evaluation metrics, the chi square technique has the best competence in term of sensitiveness while the rough set genetic algorithm reduct is good at specificity and accuracy. In the next step, further analysis will be conducted on complex dataset such as time series data set.

Item Type:	Conference or Workshop Item (Paper)
Additional Information:	INSPEC Accession Number: 14882158
Uncontrolled Keywords:	artificial immune system; danger theory; dendrite cell algorithm; feature selection; signal selection
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Divisions:	School of Computing
Depositing User:	Dr. Mohamad Farhan Mohamad Mohsin
Date Deposited:	17 May 2015 03:20
Last Modified:	22 May 2016 07:26
URI:	https://repo.uum.edu.my/id/eprint/14118

Actions (login required)

View Item