Why not just count the number of times a player can attack on a computer? I mean, you obviously need to check if it is within range, so why not check if it is within range, and if there is one, increase the counter could_have_attacked. Then, if he really attacks, increase the counter attacked. Then you can compare how often a human player attacks, how often he has the opportunity.
Alternatively, compare how often it is in range, how often it is not in range, or, as a rule, it tracks the average distance between players. More aggressive players are likely to be much closer to their opponents than to passive players.
, , . , , , .