Bubble sort can "stop early" should the array already be sorted. Other algorithms have their own advantages or applications, but none of the algorithms mentioned in your question - other than bubble sort - can "stop early".
Therefore for any input Insertion & Selection runtime will be $\Omega (n^2) $, and Mergesort $\Omega(nlog(n))$, so the order of elements doesn't matter at all.
$\bullet $ If you perform bubble sort on an already sorted array, the algorithm can stop when not performing any swaps, and run at $O(n)$
$\bullet $ Merge sort (to which you refer as 'fusion sort' from some reason) is the only algorithm that requires $O(nlog(n))$ rather than $O(n^2)$
$\bullet $ Insertion sort is useful when receiving the array online (one element at a time) and maintaining it sorted. Its noteworthy that since you maintain a sorted array, new elements can be inserted via binary search, which make the insertion sort perform $O(nlog(n))$ comparisons, but still cost $O(n^2)$ due to moving elements.
$\bullet $ Selection sort is very easy to implement and intuitive to explain