Jx-MLT : A Machine Learning Toolbox for Classification

---

"Toward Talent Scientist: Sharing and Learning Together" --- Jingwei Too

---

• This toolbox contains 8 widely used machine learning algorithms

• The A_Main file provides the examples of how to use these methods on benchmark dataset

• Main goals of this toolbox are:

  • Sharing knowledge on machine learning works
  • Helping others in machine learning projects

Usage

The main function jml is used to perform the classification. You may switch the algorithm by simply changes the 'da' to other abbreviations

• If you wish to use discriminate analysis ( DA ) classifier then you may write

ML = jml('da',feat,label,opts); 

• If you want to use naive bayes ( NB ) classifier then you may write

ML = jml('nb',feat,label,opts); 

Input

• feat : feature vector matrix ( Instance x Features )
• label : label matrix ( Instance x 1 )
• opts : parameter settings
  • tf : choose either hold-out / k-fold / leave-one-out
  • ho : ratio of testing data in hold-out validation
  • kfold : number of folds in k-fold cross-validation

Output

• ML : Machine learning model ( It contains several results )
  • acc : classification accuracy
  • con : confusion matrix
  • t : computational time (s)

How to choose the validation scheme?

There are three types of performance validations. These validation strategies are listed as following. However, if you do not select any of them, then the algorithm will automatically perform 10-fold cross-validation as default.

• Hold-out validation

opts.tf    = 1;
opts.ho    = 0.3;   % 30% data for testing 

• K-fold cross-validation

opts.tf    = 2
opts.kfold = 10;    % 10-fold cross-validation

• Leave-one-out validation

opts.tf    = 3 

Example 1 : K-nearest neighbor ( KNN ) with k-fold cross-validation

% Parameter settings
opts.tf    = 2;     
opts.kfold = 10;    
opts.k     = 3;     % k-value in KNN

% Load data
load iris.mat;

% Classification
ML = jml('knn',feat,label,opts);

% Accuracy
accuracy = ML.acc; 

% Confusion matrix
confmat  = ML.con;

Example 2 : Multi-class support vector machine ( MSVM ) with hold-out validation

% Parameter settings
opts.tf    = 1;     
opts.ho    = 0.3;       
opts.fun   = 'r';     % radial basis kernel function in SVM

% Load data
load iris.mat;

% Classification
ML = jml('msvm',feat,label,opts);

% Accuracy
accuracy = ML.acc; 

% Confusion matrix
confmat  = ML.con;

Example 3 : Decision Tree ( DT ) with leave-one-out validation

% Parameter settings
opts.tf     = 3;          
opts.nSplit = 50;    % number of split in DT 

% Load data
load iris.mat;

% Classification
ML = jml('dt',feat,label,opts);

% Accuracy
accuracy = ML.acc; 

% Confusion matrix
confmat  = ML.con;

Requirement

• MATLAB 2014 or above
• Statistics and Machine Learning Toolbox

List of available machine learning methods

• Click on the name of algorithm to check the parameters
• Use the opts to set the specific parameters

No.	Abbreviation	Name	Support
09	'gmm'	Gaussian Mixture Model	Multi-class
08	'knn'	K-nearest Neighbor	Multi-class
07	'msvm'	Multi-class Support Vector Machine	Multi-class
06	'svm'	Support Vector Machine	Binary class
05	'dt'	Decision Tree	Multi-class
04	'da'	Discriminate Analysis Classifier	Multi-class
03	'nb'	Naive Bayes	Multi-class
02	'rf'	Random Forest	Multi-class
01	'et'	Ensemble Tree	Multi-class