Volume 53, Number 1, January–March 2019
|Page(s)||269 - 288|
|Published online||15 February 2019|
Gene selection via BPSO and Backward generation for cancer classification
Laboratory of Mathematics, Computing & Applications-Information Security, Faculty of Sciences Mohammed V University in Rabat, BP1014 Rabat, Morocco
* Corresponding author: firstname.lastname@example.org
Accepted: 7 July 2018
Gene expression data (DNA microarray) enable researchers to simultaneously measure the levels of expression of several thousand genes. These levels of expression are very important in the classification of different types of tumors. In this work, we are interested in gene selection, which is an essential step in the data pre-processing for cancer classification. This selection makes it possible to represent a small subset of genes from a large set, and to eliminate the redundant, irrelevant or noisy genes. The combinatorial nature of the selection problem requires the development of specific techniques such as filters and Wrappers, or hybrids combining several optimization processes. In this context, we propose two hybrid approaches (RBPSO-1NN and FBPSO-SVM) for the gene selection problem, based on the combination of the filter methods (the Fisher criterion and the ReliefF algorithm), the BPSO metaheuristic algorithms and the Backward algorithm using the classifiers (SVM and 1NN) for the evaluation of the relevance of the candidate subsets. In order to verify the performance of our methods, we have tested them on eight well-known microarray datasets of high dimensions varying from 2308 to 11225 genes. The experiments carried out on the different datasets show that our methods prove to be very competitive with the existing works.
Mathematics Subject Classification: 62F07 / 62H30 / 62P10 / 68T20 / 90C06
Key words: Gene selection / cancer classification / BPSO / backward generation / SVM / 1NN / ReliefF / Fisher criterion / DNA microarray
© EDP Sciences, ROADEF, SMAI 2019
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.