Confusion Matrix Results Issue

I need to determine the misclassified rate of a machine learning algorithm. However, when I use the confusion function over the actual dataset and the predicted dataset (using the algorithm) the error rate is 0 whereas when I iterate through each element and compare them the error rate is 33.3%. What is wrong with the confusion matrix?

outputs         = [1, 1, 1, 100, 10, 100];
predictedOutput = [1, 1, 1,  10,100, 100];

[c,cm] = confusion(outputs,predictedOutput);

fprintf('Percentage Correct Classification   : %f%%\n', 100*(1-c));

counter =0;
for i = 1: size (predictedOutput,2)
  if (predictedOutput(1,i) ~= outputs(1,i))
       counter = counter + 1;
  end
end

fprintf('Percentage Correct Classification : %f%%\n', 100*(1-(counter/(size(predictedOutput,2)))));

Answer 1

[Shoaibur Rahman]

The input arguments of confusion, (in this case, outputs and predictedOutput) should be in range of [0 1]. So, instead of 100 and 10, use 1 and 0, for example.

outputs         = [1, 1, 1, 1, 0, 1];
predictedOutput = [1, 1, 1, 0, 1, 1];
[c,cm] = confusion(outputs,predictedOutput);
fprintf('Percentage Correct Classification   : %f%%\n', 100*(1-c));
counter =0;
for i = 1: size (predictedOutput,2)
  if (predictedOutput(1,i) ~= outputs(1,i))
       counter = counter + 1;
  end
end
fprintf('Percentage Correct Classification : %f%%\n', 100*(1-(counter/(size(predictedOutput,2)))));

[Mr X]

I to classify the dataset into 3 different classes so I cannot use 1 and 0 only. Does this mean that I need to create three rows to represent each class and then select it by setting the element to 1?

[Shoaibur Rahman]

That could be one way, given it serves your purpose. Another way is to analyze the confusion matrix (user your original outputs,predictedOutput).

cmat = confusionmat(outputs,predictedOutput);

This will allow you to determine the misclassification for each group separately.