The issue of representing 0 and 1 in a feed-forward network with 2 output neurons can be resolved by representing 0 and 1 as vectors of size 2, that is, 0 -> [1,0] and 1-> [0,1].
As vectors of size 2 can take upto 4 output values, we can easily represent by saying the 0th output neuron will "light up" when the output is 0 (no defect) and 1st output neuron will light up when the output is 1 (defect).
Some boilerplate code you can use to perform this functionality:
filename = 'software_defects.csv';
data = readtable(filename);
inputs = data{:, 1:36}';
targets = data{:, 37};
% 0 -> [1; 0] (non-defective)
% 1 -> [0; 1] (defective)
numSamples = length(targets);
targetsNN = zeros(2, numSamples);
for i = 1:numSamples
if targets(i) == 0
targetsNN(:, i) = [1; 0];
else
targetsNN(:, i) = [0; 1];
end
end
net = feedforwardnet(10); % 10 hidden neurons, adjust as needed
net = configure(net, inputs, targetsNN);
net = train(net, inputs, targetsNN);
predictions = net(inputs);
Hope this helps!
