why my models testing accuracy gets worse

I have written my code below and dataset is also attached. After applyng the 10 fold cross-validation, the testing accuracy gets worse while there is no problem in training accuracy. Please help me to resovle this issue.

data1=xlsread('waveform.csv');

data=data1(:,1:end);

groups=data1(:,end);

Fold=10;

indices = crossvalind('Kfold',length(groups),Fold); % create indices of 10 fold cross-validation, Group is the vector containing the class label for each obsevation

for i =1:Fold % create train and test sets

display(['cross validation, folds' num2str(i)])

testy = (indices == i);

trainy = (~testy);

TrainInputData=data(trainy,:);

TrainOutputData=groups(trainy,:);

TestInputData=data(testy,:);

TestOutputData=groups(testy,:);

%

% set the parameters

%

% regularization parameter: [C1, C2, C3] for each layer respectively

% kernel parameters: [SIG1, SIG2, SIG3] for each layer respectively

C1 = 1; C2 = 1; C3 = 1;

SIG1 = 1; SIG2 = 1; SIG3 = 1;

[TrainingTime, TestingTime, TrainingAccuracy, TestingAccuracy] = ...

MLKELM([TrainInputData TrainOutputData], [TestInputData TestOutputData], 1, [C1, C2, C3], 'RBF_kernel', [SIG1, SIG2, SIG3], 3)

testing_Accuracy_f(i)=TestingAccuracy;% keep testing acc for each fold

end

mean=sum(testing_Accuracy_f)/length(testing_Accuracy_f);

StandDevx = sqrt(sum((testing_Accuracy_f-mean).^2)/(length(testing_Accuracy_f)-1));