Ant Colony Optimization for Feature Selection

Introduction

• This toolbox offers ant colony optimization method ( ACO )
• The Main file illustrates the example of how ACO 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 ants
• max_Iter : maximum number of iterations
• alpha : coefficient control tau
• beta : coefficient control eta
• tau : initial tau
• eta : initial eta
• rho : pheromone

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;
tau      = 1;
alpha    = 1; 
rho      = 0.2;
beta     = 0.1; 
eta      = 1;

% Ant Colony Optimization 
[sFeat,Nf,Sf,curve] = jACO(feat,label,N,max_Iter,tau,eta,alpha,beta,rho,HO);

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

Requirement

• MATLAB 2014 or above
• Statistics and Machine Learning Toolbox