Getting Accuracy from a Confusion Chart

Hello,

I can't seem to find the right documentation or i i misunderstand something but i want to try to find the accuracy of a confusion chart using "confusionchart". I know the algorithms and etc. and i can easily find the algorthms needed but i can't seem to get what i need. code and image included. The code is from in class lectures

clear  
close all 
clc

% Load training and test data using |imageDatastore|.
Training_Dir   = fullfile(toolboxdir('vision'),'visiondata','digits','synthetic');
Testing_Dir = fullfile(toolboxdir('vision'),'visiondata','digits','handwritten');

% imageDatastore recursively scans the directory tree containing the
% 
trainingSet = imageDatastore(Training_Dir,'IncludeSubfolders',true,'LabelSource','foldernames');
testingSet     = imageDatastore(Testing_Dir,'IncludeSubfolders',true,'LabelSource','foldernames');
%% Show a few of the training and test images
close all 
figure('Name','Training Image Examples','NumberTitle','off');
movegui("northwest")
subplot(3,3,1);
imshow(trainingSet.Files{12});

subplot(3,3,2);
imshow(trainingSet.Files{315});

subplot(3,3,3);
imshow(trainingSet.Files{810});
subplot(3,3,4);
imshow(trainingSet.Files{1000});
subplot(3,3,5);
imshow(trainingSet.Files{500});

subplot(3,3,6);
imshow(testingSet.Files{10});

subplot(3,3,7);
imshow(testingSet.Files{35});

subplot(3,3,8);
imshow(testingSet.Files{80});
subplot(3,3,9);
imshow(testingSet.Files{105});

%% Show pre-processing  and Feature Extraction results
exTestImage = readimage(testingSet,12);
Gray_IM=im2gray(exTestImage);
Binary_IM = imbinarize(Gray_IM);

figure('Name','Preprocessed Image','NumberTitle','off');
movegui("north")

subplot(1,3,1)
imshow(exTestImage)
subplot(1,3,2)
imshow(Gray_IM)
subplot(1,3,3)
imshow(Binary_IM)

%%  Extract HOG features from image

[img, imgInfo] = readimage(trainingSet, 800);
imgInfo
cellSize = [2 2];
[hog_4x4, vis4x4] = extractHOGFeatures(img,'CellSize',cellSize);
figure('Name','Histogram of Gradients','NumberTitle','off')
movegui("northeast")

plot(vis4x4); 
figure('Name','Training Image Example with HOG','NumberTitle','off');

movegui("southwest")

imshow(img, 'InitialMagnification',2000)
hold on
plot(vis4x4,'color', 'b')

hold off

%% Extracting HOG Features

% HOG features from each image in training set.


hogFeatureSize = length(hog_4x4);
numImages = numel(trainingSet.Files);
trainingFeatures = zeros(numImages,hogFeatureSize,'single');
for i = 1:numImages
    img = readimage(trainingSet,i);
    
    img = im2gray(img); %convertign the specified truecolor The RGB Images to a grayscale intensity image 
    
   
    img = imbinarize(img);   %  pre-processing step to conver the Imges into black & white
    
    trainingFeatures(i, :) = extractHOGFeatures(img,'CellSize',cellSize);  
end

% similar procedure will be used  with testing set.

numImages_test = numel(testingSet.Files);

testingFeatures = zeros(numImages_test,hogFeatureSize,'single');
for i = 1:numImages_test
    img = readimage(testingSet,i);
    
    img = im2gray(img); %convertign the specified truecolor The RGB Images to a grayscale intensity image 
    
   
    img = imbinarize(img);   %  pre-processing step to conver the Imges into black & white
    
    testingFeatures(i, :) = extractHOGFeatures(img,'CellSize',cellSize);  
end
%% Train and Testing a Decision Tree Classifier


% Get labels for each image.
trainingLabels = trainingSet.Labels;
tree = fitctree(trainingFeatures,trainingLabels);

% Evaluating the the Classifier
% Get labels for each image.
testLabels = testingSet.Labels;

predictedLabels = predict(tree, testingFeatures);

% Tabulate the results using a confusion matrix.
 figure("Name","Confusion Matrix",'NumberTitle','off')
 movegui("south")
cm = confusionchart(testLabels,predictedLabels,'RowSummary','row-normalized','ColumnSummary','column-normalized');



%% Testing the performance of Trained model on some testing images 

M=20;
predicted_Image = predict(tree, testingFeatures(M,:));
figure("Name","Predicted Digit", "NumberTitle","off")
imshow(testingSet.Files{M}, 'InitialMagnification',800);
caption = sprintf('Predicted Digit is: %c', predicted_Image );
title(caption, 'FontSize', 18);

Answer 1

[Sai Pavan]

Hi Richard,

I understand that you are trying to get the accuracy of the model from the confusion chart.

Please follow the below workflow to get the accuracy of the model:

• Generate the confusion chart using the “confusionchart” function which requires the true labels and predicted labels as inputs.
• Access the confusion matrix from the “chart” object using the “NormalizedValues” property. 
• Calculate the accuracy using the elements of the confusion matrix. The accuracy is the sum of the diagonal elements divided by the sum of all elements.

Please refer to the below code snippet to get the accuracy of the model from the confusion chart:

chart = confusionchart(testLabels,predictedLabels); % Generate confusion chart
confusionMatrix = chart.NormalizedValues; % Get the normalized confusion matrix
accuracy = sum(diag(confusionMatrix)) / sum(confusionMatrix(:)); % Calculate accuracy

Hope it helps.

Regards,

Sai Pavan