As per the documentation of "fitcecoc" given here https://www.mathworks.com/help/stats/fitcecoc.html#bue3oc9-2, you can see the input table takes in a 2D data, which is basically, numObservations x numFeatures. This means each observation (or sample) is a row, and each feature is a column. If your data is inherently multi-dimensional (like images or matrices), you need to flatten these into a single row to fit this format.
I am assuming your data is a matrix, and hence it will need to be flattened out.
Following that, it is a simple implementation of an "fitcecoc" classifier. Here is some boilerplate code:
% Assume 'data' is your cell array where each cell is a 20x120 matrix
% Assume 'labels' is a vector containing the label for each matrix
numObservations = numel(data);
reshapedData = zeros(numObservations, 20 * 120);
for i = 1:numObservations
reshapedData(i, :) = reshape(data{i}, 1, []);
end
% Train the SVM using fitcecoc
% 'labels' should be a column vector with the same number of rows as reshapedData
svmModel = fitcecoc(reshapedData, labels);
Hope this helps!
