1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
|
function [selectedFeatures] = FCBF(featureMatrix,classColumn,threshold)
%function [selectedFeatures] = FCBF(featureMatrix,classColumn,threshold)
%
%Performs feature selection using the FCBF measure by Yu and Liu 2004.
%
%Instead of selecting a fixed number of features it provides a relevancy threshold and selects all
%features which score above that and are not redundant
%
% The license is in the license.txt provided.
numFeatures = size(featureMatrix,2);
classScore = zeros(numFeatures,1);
for i = 1:numFeatures
classScore(i) = SU(featureMatrix(:,i),classColumn);
end
[classScore indexScore] = sort(classScore,1,'descend');
indexScore = indexScore(classScore > threshold);
classScore = classScore(classScore > threshold);
if ~isempty(indexScore)
curPosition = 1;
else
curPosition = 0;
end
while curPosition <= length(indexScore)
j = curPosition + 1;
curFeature = indexScore(curPosition);
while j <= length(indexScore)
scoreij = SU(featureMatrix(:,curFeature),featureMatrix(:,indexScore(j)));
if scoreij > classScore(j)
indexScore(j) = [];
classScore(j) = [];
else
j = j + 1;
end
end
curPosition = curPosition + 1;
end
selectedFeatures = indexScore;
end
function [score] = SU(firstVector,secondVector)
%function [score] = SU(firstVector,secondVector)
%
%calculates SU = 2 * (I(X;Y)/(H(X) + H(Y)))
hX = h(firstVector);
hY = h(secondVector);
iXY = mi(firstVector,secondVector);
score = (2 * iXY) / (hX + hY);
end
|
