Atom Search Optimization for Feature Selection

Introduction

• This toolbox offers an Atom Search Optimization ( ASO ) method
• The Main file illustrates the example of how ASO can solve the feature selection problem using benchmark data-set.

Input

• feat : feature vector ( Instances x Features )
• label : label vector ( Instances x 1 )
• N : number of atoms
• max_Iter : maximum number of iterations
• alpha : depth weight
• beta : multiplier weight

Output

• sFeat : selected features
• Sf : selected feature index
• Nf : number of selected features
• curve : convergence curve

Example

% Benchmark data set 
load ionosphere.mat;  

% Set 20% data as validation set
ho = 0.2; 
% Hold-out method
HO = cvpartition(label,'HoldOut',ho);

% Parameter setting
N        = 10; 
max_Iter = 100; 
alpha    = 50; 
beta     = 0.2;
% Atom Search Optimization
[sFeat,Sf,Nf,curve] = jASO(feat,label,N,max_Iter,alpha,beta,HO);

% Plot convergence curve
plot(1:max_Iter,curve);
xlabel('Number of iterations');
ylabel('Fitness Value');
title('ASO'); grid on;

Requirement

• MATLAB 2014 or above
• Statistics and Machine Learning Toolbox

Cite As

@article{too2020chaotic,
  title={Chaotic Atom Search Optimization for Feature Selection},
  author={Too, Jingwei and Abdullah, Abdul Rahim},
  journal={Arabian Journal for Science and Engineering},
  pages={1--17},
  year={2020},
  publisher={Springer}
}