correct and incorrect predictors

Dear community members, i am stuck in a problem. I tried to search for the solution but i could not find. I want to get the input data (predictors) of an incorrect predicted class. How can i do that? i am just able to get the incorrect classes in the confusion chart but i need to find the input data of them.

for example lets take this example:

load satdata;
pt = cvpartition(satClass,'holdout',0.3); 
predTrain = satData(training(pt),:);
classTrain = satClass(training(pt));
predValid = satData(test(pt),:);
classValid = satClass(test(pt));
knnClassifier = fitcknn(predTrain,classTrain,'Numneighbors',5);
yPred = predict(knnClassifier,predValid);
[c,lbls] = confusionmat(yPred,classValid);

Here i can only see the classes in yPred but i cant see the input data (predictors) of those classes. I hope i am clear to my question.

[Walter Roberson]

Unfortunately I cannot tell from here what class(classTrain) is, and so what class(yPred) is.

[Saeed Magsi]

load fisheriris
pt = cvpartition(species,'holdout',0.3); 
predTrain = meas(training(pt),:);
classTrain = species(training(pt));
predValid = meas(test(pt),:);
classValid = species(test(pt));
knnClassifier = fitcknn(predTrain,classTrain,'Numneighbors',5);
yPred = predict(knnClassifier,predValid);
[c,lbls] = confusionmat(yPred,classValid);

Sorry for uploading the wrong code. I have now written a matlab documentation example code for knn classifier.

[Saeed Magsi]

I have now attached the excel files for yPred and classTrain. If you see in row 23 of yPred it has incorrect prediction of "virginica". Now i want to see its input data (predictors) which i dont know how can this be done? I need your help in this. Thank you once again.

Answer 1

[Dinesh]

Hi Saeed!

One simple way to find the incorrect prediction is to iterate over the predictions and store the indices that are not equal in yPred and classValid. Using that indices array we can get the input data predictors for which the predictions were wrong.

indexes = [];

for i = 1:length(yPred) 
    % Compare values and store index if they don't match 
    if ~strcmp(yPred{i}, classValid{i}) 
        indexes(end+1) = i; 
    end 
end 
data = predValid(indexes, :); 
disp(data); 

You can do this without for loop also. Use the 'cellfun' fuction

res = cellfun(@isequal, yPred, classValid); 
% Find the indexes of non-matching elements 
indexes = find(~res); 
indexes 
data = predValid(indexes, :); 
disp(data); 

Please refer to the following MATLAB documentation for more details and examples to use cellfun

https://www.mathworks.com/help/matlab/ref/cellfun.html

Hope this helps!

Thank you.