A thermodynamic algorithm for feature selection
Document typeConference report
PublisherThomson Editores Spain
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder
The main purpose of Feature Selection (FS) is to find a reduced subset of attributes from a data set described by a feature set. This implies a search process in the space of possible solutions, trying to optimize an objective function. This work introduces TAFS, a Thermodynamic Annealing Feature Selection algorithm. Given a suitable objective function, TAFS uses a special-purpose implementation of simulated annealing to find a good subset of attributes that maximizes this objective function. A distinctive characteristic of TAFS over other search algorithms for feature subset selection is its probabilistic capability to accept momentarily worse solutions. TAFS has been evaluated against one of the most robust and reliable algorithm, the Sequential Forward Floating Search method (SFFS). Our experimental results show that TAFS achieves significant improvements over SFFS in the objective function for classification tasks with a reasonable reduction in subset size.
CitationBelanche, L.; González, F. A thermodynamic algorithm for feature selection. A: Simposio de Inteligencia Computacional. "Actas del Simposio de Inteligencia Computacional, SICO'2007". Thomson Editores Spain, 2007, p. 57-64.