Here's the most concise (and working) way to plot your data:
figure(1);
hold all
plot(train1(:,1),train1(:,2),'o')
plot(train2(:,1),train2(:,2),'x')
plot(train3(:,1),train3(:,2),'s')
Here's an example that does what you want in a robust and modular manner. You can easily add classes or modify the figure output.
data = [0.53,0.17,2;0.78,0.60,3;0.93,0.26,1;0.13,0.65,2;0.57,0.69,1;...
0.47,0.75,3;0.010,0.45,1;0.34,0.080,3;0.16,0.23,3;0.79,0.91,3;...
0.31,0.15,1;0.53,0.83,2];
categories = [1,2,3];
symbols = {'s','<','o','d','v','+','x','*'};
figure;
hold all
for loopj = 1:length(categories)
t = data(:,3) == categories(loopj);
train = data(t,:);
label = strcat('Class ',num2str(categories(loopj)));
plot(train(:,1),train(:,2),symbols{loopj},'DisplayName',label,'LineWidth',1.3)
end
lg = legend('show');
lg.Location = 'best';
Use hold all
to write on the figure without erasing the previous axi and let Matlab pick line colors.
In any case, you need to manually define the different symbols and each plot command comes with one unique type of line and markers.