We know that the four-dimensional scalar product is invariant under coordinate transformation, hence the space-time interval and proper time is also invariant.
Correct. Space-time interval in $(-,+,+,+)$ signature is $ds^2 = -c^2dt^2+dx^2+dy^2+dz^2$ and is Lorentz invariant, and by definition, proper time $d\tau = \frac{\sqrt{-ds^2}}{c}$ is also invariant.
Since the 4-velocity is given by space-time interval divided by proper time. Then why 4-velocity is not invariant?
No, 4-velocity is $\vec u = \frac{\vec {ds}}{d\tau} = (\gamma, \gamma \frac{\vec v}{c})c$. Now this cannot be invariant for all cases, precisely because of time-dilation. It can be in case of time-dilation factor being one or equivalently $v = 0$ where the observer's frame of reference is not moving at all. This should be true, just as expected.