Particle Swarm Optimization for Feature Selection

Introduction

• This toolbox offers a Particle Swarm Optimization (PSO) method
• The Main file illustrates the example of how PSO 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 particles
• max_Iter : maximum number of iterations
• c1 : Cognitive factor
• c2 : Social factor
• w : Inertia 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;
c1       = 2;     % cognitive factor
c2       = 2;     % social factor
w        = 1;     % inertia weight

% Particle Swarm Optimization
[sFeat,Sf,Nf,curve] = jPSO(feat,label,N,max_Iter,c1,c2,w,HO);

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

Requirement

• MATLAB 2014 or above
• Statistics and Machine Learning Toolbox