To eliminate ISI, what's required is a Nyquist filter. The raised-cosine and sinc filters are two examples of Nyquist filters, so either one will, in theory, eliminate ISI.
The Gaussian filter is not a Nyquist filter. You know this because its impulse response is never zero, while a Nyquist filter must have its impulse response go to zero at all times $t = n T_0$. A Gaussian filter might be used in an MMF system because the MMF channel itself has a Gaussian response, so that a Gaussian filter is a channel-matching filter, optimizing received SNR, in these systems.
In fiber optics what is the most suitable pulse shaping filter that can be used to design the desired pulse shape to mitigate the Inter-symbol Interference (ISI)
Any Nyquist filter may be used to minimize ISI.
Practically, building such filters for fiber optic modulation frequencies is difficult, and we simply use a bandwidth-limited receiver that more-or-less matches the desired system bandwidth. Increasing the receiver bandwidth generally improves ISI but degrades received noise, so the receiver is designed to optimize the net effect on bit error rate.