We want to use Gaussian elimination to normalize this parity check matrix H into normalized form [P I].
Since H is a 15x20 matrix, P would be a 15x5 matrix and I would be a 15x15 matrix.
However, observe that the last 15 entries of the first row in H are all zeroes.
Thus, it is impossible to normalize H using Gaussian elimination alone.
We might try to use a slightly different approach to normalize H, namely the method suggested in this StackExchange answer https://math.stackexchange.com/questions/2818445/how-do-i-find-parity-check-matrix-if-generator-matrix-cant-be-written-in-standa :
Convert H to row-echolon form
Perform column-swaps to produce normalized from [P I]
Create the normalized generator matrix G = [I PT]
Undo the column-swaps from step 2 onto the generator matrix G
However, for the matrix H given in your example, this approach does also not work because the rows are linearly dependent and the row-echolon form has all-zeroes as its last two rows.
The issue of how to generate a reasonable generator matrix G even when the parity check matrix H has linear dependencies is discussed here: