function y = common_by_row(x)
tmp=[];
y=[];
z=[];
k=1;
[r,c]=size(x);
for i=1:r
for j=1:c
if x(i,j)~=NaN
if ismember(x(i,j),z)==0
z(k)=x(i,j);
k=k+1;
end
end
end
end
cnt=0;
for k=1:length(z)
for i=1:r
for j=1:c
if x(i,j)==z(k)
cnt=cnt+1;
end
end
end
tmp(k)=cnt;
cnt=0;
end
m=1;
for i=1:length(z)
if tmp(i)>0.5*r
y(m)=z(i);
m=m+1;
end
end
y=sort(y);
end

I know I have adapted code according to test cases. But was stuck in this one.

Don't know why an empty matrix 1-by-0 is not equal to the empty matrix [].

[] is 0-by-0. mtimes(ones(1,0), ones(0,1)) equals 0, mtimes([], ones(0,1)) is empty 0-by-1. This is one of the reasons.

I used exactly the same code as for problem 67.

Nice! But I think you made a mistake in if length(m)==size(x,1), according to the instruction, a number that appears more than half of total rows would be okay.

My solution fails on the magic, the wilkonson and the [1;2] inputs.
However, on my own computer each input works as expected.
Anyone has a hint on where I am going wrong?

Found it.
Due to the way the test-suite is set-up (using 'isequal'), MATLAB makes a difference between the values "[]" (size [0 0]) and "Empty matrix: 1-by-0" (size [1 0]), even though they are both considered empty.