during the classification of malayalam short vowel classification ,i'm getting the accuracy of 20%.How can i increase the accuracy,

the code is given below.How can i increase the accuracy with respect to the vowel classification.

clear; clc; DatabasePath='D:\Education\3sem\project\newspeech(mv)\mv_database'; NumberOfVowels=5; NumberOfSamplesPerVowel=250; k1=1; k2=1; %x=zeros(500,3500); for index1=1 : NumberOfVowels FolderName=sprintf('%s\\%d',DatabasePath,index1); for index2=1 : NumberOfSamplesPerVowel sn=wavread(sprintf('%s\\%d.wav',FolderName,index2)); if index2<=125 TrainData(k1,:)=sn(1:1100); k1=k1+1; else TestData(k2,:)=sn(1:1100); k2=k2+1; end %sound(x1,fs1); end end [P,N]=size(TrainData);

%finding the minimum and maximum min1=min(TrainData(1:P,:)); max1=max(TrainData(1:P,:));

% Finding the normalizes data

for i=1:N for j=1:P Train_norm(j,i)=(TrainData(j,i)-min1(i))/(max1(i)-min1(i)); end end min2=min(TestData(1:P,:)); max2=max(TestData(1:P,:));

% Finding the normalizes data

for i=1:N for j=1:P Test_norm(j,i)=(TestData(j,i)-min2(i))/(max2(i)-min2(i)); end end

m1=mean(Train_norm(1:125,:)); m2=mean(Train_norm(126:250,:)); m3=mean(Train_norm(251:375,:)); m4=mean(Train_norm(376:500,:)); m5=mean(Train_norm(501:625,:));

% within class scatter matrix of class 1 Sw1=zeros(1100); for i=1:125 Diff=Train_norm(i,:)-m1; Sw1=Sw1+(Diff'*Diff); end

% within class scatter matrix of class 2 Sw2=zeros(1100); for i=126:250 Diff=Train_norm(i,:)-m2; Sw2=Sw2+(Diff'*Diff); end % within class scatter matrix of class 3 Sw3=zeros(1100); for i=251:375 Diff=Train_norm(i,:)-m3; Sw3=Sw3+(Diff'*Diff); end % within class scatter matrix of class 4 Sw4=zeros(1100); for i=376:500 Diff=Train_norm(i,:)-m4; Sw4=Sw4+(Diff'*Diff); end % within class scatter matrix of class 5 Sw5=zeros(1100); for i=501:625 Diff=Train_norm(i,:)-m5; Sw5=Sw5+(Diff'*Diff); end

% sum of within class scatter matrix Sw=Sw1+Sw2+Sw3+Sw4+Sw5;

% total mean m=(m1+m2+m3+m4+m5)/5;

% between class scatter matrix Sb1=125*(m1-m)'*(m1-m); Sb2=125*(m2-m)'*(m2-m); Sb3=125*(m3-m)'*(m3-m); Sb4=125*(m4-m)'*(m4-m); Sb5=125*(m5-m)'*(m5-m); Sb=Sb1+Sb2+Sb3+Sb4+Sb5;

% Computation of eigenvectors and eigen value [Evect,Eval]=eig(inv(Sw)*Sb); [Eval,ind]=sort(diag(Eval),'descend'); no_of_dim=4; EV1=Evect(:,ind(1:no_of_dim)); projtrain_data=Train_norm*EV1; projtest_data=Test_norm*EV1; projtest_data=TestData*EV1; class(1:125)=1; class(126:250)=2; class(251:375)=3; class(376:500)=4; class(501:625)=5; %gscatter(TrainData(:,1),TrainData(:,2),class); class=class'; oclass=knnclassify(projtest_data,projtrain_data,class,15); %figure,gscatter(TestData(:,1),TestData(:,2),oclass,'mc'); figure, plot(oclass(1:125,1),'*r'); hold on; plot(oclass(126:250,1),'+g'); hold on; plot(oclass(251:375,1),'^b'); hold off; c=confusionmat(class,oclass) correct=diag(c); accuracy=(sum(correct)/625)*100; display(accuracy);