The available gear in game is known and finite, provided you can eventually obtain it. The number of bosses and unique elements of each boss is known and finite. A small amount of information is known about the gear setup and stats of tanks who have successfully tanked each fight, barring other variables like the guild's overall raiding prowess.
The basis question is, for any given encounter, barring gimmicks and such, would is the ideal gear set (in terms of avoidance and HP/block/AC) to beat it. Yes, fights can be beaten with the wrong types of gear sets, but that is not my point. Theoretically the highest number of successful encounters were beaten with the correct gear more often than those with incorrect gear. There is some balance of avoidance and EH. There are some minimums, sure, but these minimums are both actual minimums due to the boss attack damage per hit or specials, and more realistic minimums in terms of what healers can handle and other gimmicks.
I think the next step is to come up with a dataset I could use. I am specifically not too worried about spell damage and threat output at the moment, this is for surviving the fight to beat it, as if there were no DPS raiders and there was no DPS requirement (i.e. your healers have to keep you alive for 6 minutes, then the boss automatically dies). Right now the variables I have in mind are:
* Encounter - name of boss etc
* Zone - if encounter is too specific
* Dodge
* Parry
* Defense
* Block Rating
* Block Value
* Actual Buffed Health
* Armor Count
* Success - whether the encounter was won or lost. true/false is good enough, although our target attribute could be more specific if necessary.
* Other variables I need to think more...
Essentially, you should be able to develop a simple classification model, kinda like a regression (like CART or CHAID). After making sure the data is not biased and such, tanks could input their variables into the model for any given encounter and on the outcome get their probability of success. They could also play around and see what gear swaps would improve or worsen the outcome.
It is kind of like the simulator, but I am trying to actually create a model, not just see results. The model would ideally be trained on actual data pulled over the course of a few months. I just need to figure out a way to build this. I almost need a thread where each time you tank a boss (win or lose), you post the bosses name and your values for each of these variables (what was your dodge, what was your parry, etc). That would serve as a great starting point. Although I doubt this community would be up for it, and I don't want to abuse the forums.
Maybe I will make a web service that collects it into a DB and then link it to there. Everyone can go type in their values (or maybe just link to their armory) and the encounter. I just hope it would not get publically abused.
_________________
