It depends on the molecular mechanism that resulted in such deletion. There could be a simple random nucleotide addition, insertion of palindromic sequence (see this example for physiological TCR recombination), trinucleotide repeats, etc. Here is another nice paper on the paper on the diverse composition of translocation junctions including inserted/deleted nucleotides.

But I would suggest you first test your algorithm on fully random data and the select the cases when it fails and study them. If there are no specific purposes of optimization or some suggestions how utilize some characteristic features of mutations to improve precision/recall for a specific mechanism, I would suggest starting with random (uniform) data.

Moreover, I believe that in your task it is more important to distinguish your indels from indel errors, e.g. homopolymer indel errors of Illumina (See here) / indel error plethora generated by 454 (See here). So I would recommend starting with analyzing patterns of those.