breast_cancer_wisconsin_radom.m

使用高斯模型对威斯康辛州大学医学院长期乳腺癌数据进行了贝叶斯模式识别。识别率为95以上

clear;
load data.mat
load ID.mat
Total  = length(data(:,1));
VaribleNumber = length(data(1,:));
Index = [1:Total];


[TestMatrix,TrainMatrix,TestIndex,TrainIndex] = SplitMatrixRandom(data);
TrainNumber = length(TrainIndex);
TestNumber = Total - TrainNumber;

Answer = strcmp(textdata(1:Total,2),'B');

TrainNumber_B =  sum(Answer(TrainIndex));
TrainNumber_M = TrainNumber - TrainNumber_B;
P_B = TrainNumber_B / TrainNumber;
P_M = TrainNumber_M / TrainNumber;

Train_B_Matrix = TrainMatrix(Answer(TrainIndex) == 1,:);
Train_M_Matrix = TrainMatrix(Answer(TrainIndex) == 0,:);

B_MeanMatrix = mean(Train_B_Matrix);
M_MeanMatrix = mean(Train_M_Matrix);
B_ConvarianceMatrix = cov(Train_B_Matrix);
M_ConvarianceMatrix = cov(Train_M_Matrix);

%zero-one loss function
TestCondition_B = zeros(TestNumber,1);
invB_ConvarianceMatrix = inv(B_ConvarianceMatrix);
detB_ConvarianceMatrix = sqrt(det(B_ConvarianceMatrix));

TestCondition_M = zeros(TestNumber,1);
invM_ConvarianceMatrix = inv(M_ConvarianceMatrix);
detM_ConvarianceMatrix = sqrt(det(M_ConvarianceMatrix));
pi2 = (2*pi)^(VaribleNumber/2);

for i = 1:TestNumber
    TestCondition_B(i,1) = exp(-.5*(TestMatrix(i,:)-B_MeanMatrix)*invB_ConvarianceMatrix*(TestMatrix(i,:)-B_MeanMatrix)')/detB_ConvarianceMatrix/pi2;
end
for i = 1:TestNumber
    TestCondition_M(i,1) = exp(-.5*(TestMatrix(i,:)-M_MeanMatrix)*invM_ConvarianceMatrix*(TestMatrix(i,:)-M_MeanMatrix)')/detM_ConvarianceMatrix/pi2;
end

Result = zeros(TestNumber,1);
ConfusionMatrix =zeros(2,2);
Result(find(TestCondition_B ./ TestCondition_M * P_B / P_M > 1)) = 1;
ConfusionMatrix(1,1) = sum(and((Result == ones(TestNumber,1)) , (Answer(TestIndex,1) == ones(TestNumber,1))));
ConfusionMatrix(1,2) = sum(and((Result == zeros(TestNumber,1)) , (Answer(TestIndex,1) == ones(TestNumber,1))));
ConfusionMatrix(2,1) = sum(and((Result == ones(TestNumber,1)) , (Answer(TestIndex,1) == zeros(TestNumber,1))));
ConfusionMatrix(2,2) = sum(and((Result == zeros(TestNumber,1)) , (Answer(TestIndex,1) == zeros(TestNumber,1))));
ConfusionMatrix
performance = sum((Result == Answer(TestIndex,1))) / TestNumber


%%%%%%%%%%%%%%%%%%%%
tmp = TrainIndex;
TrainIndex = TestIndex;
TestIndex = tmp;
tmp = TrainMatrix;
TrainMatrix = TestMatrix;
TestMatrix = tmp;
tmp = TrainNumber;
TrainNumber = TestNumber;
TestNumber = tmp;

TrainNumber_B =  sum(Answer(TrainIndex));
TrainNumber_M = TrainNumber - TrainNumber_B;
P_B = TrainNumber_B / TrainNumber;
P_M = TrainNumber_M / TrainNumber;

Train_B_Matrix = TrainMatrix(Answer(TrainIndex) == 1,:);
Train_M_Matrix = TrainMatrix(Answer(TrainIndex) == 0,:);

B_MeanMatrix = mean(Train_B_Matrix);
M_MeanMatrix = mean(Train_M_Matrix);
B_ConvarianceMatrix = cov(Train_B_Matrix);
M_ConvarianceMatrix = cov(Train_M_Matrix);

%zero-one loss function
TestCondition_B = zeros(TestNumber,1);
invB_ConvarianceMatrix = inv(B_ConvarianceMatrix);
detB_ConvarianceMatrix = sqrt(det(B_ConvarianceMatrix));

TestCondition_M = zeros(TestNumber,1);
invM_ConvarianceMatrix = inv(M_ConvarianceMatrix);
detM_ConvarianceMatrix = sqrt(det(M_ConvarianceMatrix));
pi2 = (2*pi)^(VaribleNumber/2);

for i = 1:TestNumber
    TestCondition_B(i,1) = exp(-.5*(TestMatrix(i,:)-B_MeanMatrix)*invB_ConvarianceMatrix*(TestMatrix(i,:)-B_MeanMatrix)')/detB_ConvarianceMatrix/pi2;
end
for i = 1:TestNumber
    TestCondition_M(i,1) = exp(-.5*(TestMatrix(i,:)-M_MeanMatrix)*invM_ConvarianceMatrix*(TestMatrix(i,:)-M_MeanMatrix)')/detM_ConvarianceMatrix/pi2;
end

Result = zeros(TestNumber,1);
ConfusionMatrix =zeros(2,2);
Result(find(TestCondition_B ./ TestCondition_M * P_B / P_M > 1)) = 1;
ConfusionMatrix(1,1) = sum(and((Result == ones(TestNumber,1)) , (Answer(TestIndex,1) == ones(TestNumber,1))));
ConfusionMatrix(1,2) = sum(and((Result == zeros(TestNumber,1)) , (Answer(TestIndex,1) == ones(TestNumber,1))));
ConfusionMatrix(2,1) = sum(and((Result == ones(TestNumber,1)) , (Answer(TestIndex,1) == zeros(TestNumber,1))));
ConfusionMatrix(2,2) = sum(and((Result == zeros(TestNumber,1)) , (Answer(TestIndex,1) == zeros(TestNumber,1))));
ConfusionMatrix
performance = sum((Result == Answer(TestIndex,1))) / TestNumber