cross-validation for decision tree with matlab

data_label={'Revenu','Propriete','Credit ','Classe'};

my_data={'Eleve','Superieur','Non','C1';
'Eleve','Superieur','Oui','C2';
'Eleve','Superieur','Non','C1';
'Eleve','Inferieur','Oui','C2';
'Moyen','Superieur','Non','C1';
'Moyen','Superieur','Oui','C2';
'Moyen','Inferieur','Non','C2';
'Moyen','Inferieur','Oui','C2';
'Faible','Inferieur','Non','C3';
'Faible','Inferieur','Oui','C3'}; 
and the code for cross-validation:

indices = crossvalind('Kfold',my_data(:,end),10);
cp = classperf(my_data(:,end));
for i = 1:10
    test = (indices == i); train = ~test;
    class = classify(my_data(test,:),my_data(train,:),my_data(train,:));
    classperf(cp,class,test)
end
cp.ErrorRat

But i have this error:

Index exceeds matrix dimensions.

Error in test (line 23)
    class = classify(my_data(tests,:),my_data(train,:),data_label(train,:));

Please helppppp !!!!

Answer 1

[Saurabh]

Hi @chikilon,

The error arises because 'classify' expects numerical input, but the data provided is in a cell array of strings.

Resolution Steps:

• Convert Cell Array to Table:

data_label = {'Revenu','Propriete','Credit','Classe'};
my_data_table = cell2table(my_data, 'VariableNames', data_label);

• Seprate Features and Labels

features = my_data_table(:, 1:end-1);
labels = my_data_table.Classe;

• Convert Categorical Variables to Numeric:

features_array = table2array(features);
features_categorical = categorical(features_array);
features_numeric = dummyvar(features_categorical);

• Perform 10-Fold Cross-Validation:

indices = crossvalind('Kfold', labels, 10);
cp = classperf(labels);

for i = 1:10
    test = (indices == i);
    train = ~test;
    class = classify(features_numeric(test, :), features_numeric(train, :), labels(train));
    classperf(cp, class, test);
end

error_rate = cp.ErrorRate;

This approach ensures compatibility with the 'classify' function by providing numerical inputs and properly formatted labels.

I hope this helps.