My bad. Didn't think that through. A guestimate from me would be one to find 5 inputs (That's how many addresses I have with unspent balances rn) and 2 outputs.
It really comes down to how people use it. I tend to 'consolidate' multiple inputs from the same address from time to time. I would be interested in some statistics though (e.g. average amount of inputs and outputs per TX).
I wouldn't say it was totally random. I've once gotten a 0.0005 tx (one input two outputs) delayed 3 blocks.
And where is the math behind it? You can't just predict such a number based on your own (past trends). In either case, you'd have to choose a TX size (e.g. 500 bytes) and then try to calculate the necessary (estimated) fees required to get it included into that specific block.