Hi Merlin,
As per my understanding, you are implementing the Fuzzy K-Nearest Neighbours (FKNN) algorithm to classify your data. There seems to be an issue with calculating the class membership of each test sample, particularly in handling distance calculations and membership degree computations.
In the provided code, the use of “cumsum” for membership calculations do not align with FKNN requirements. Here is a concise snippet that demonstrates how to compute the fuzzy membership for each class:
num_classes = 5;
y_est = zeros(size(x_test, 1), num_classes);
% Calculate fuzzy membership
for i = 1:size(x_test, 1)
membership = zeros(1, num_classes);
for j = 1:k
class_label = y_train(idx(i, j));
weight = 1 / (dists(i, j) .^ (2 / (m - 1)));
membership(class_label + 1) = membership(class_label + 1) + weight;
end
y_est(i, :) = membership / sum(membership);
end
For additional insights and examples on implementing FKNN, you can refer to these resources:
- https://www.mathworks.com/matlabcentral/fileexchange/13358-fuzzy-k-nn
- https://www.mathworks.com/matlabcentral/fileexchange/21326-fuzzy-k-nn
Hope this helps!
