Join our Discord
At GGPredict we have built a system for pairwise team comparison, which gives us probabilities of win between any two teams in professional CS:GO. It is a simple system, not accounting for maps being played or whether the match is taking place on LAN or online, but the results are still significantly better than in the case of bookmaker odds and other most simple predictive systems. It looks at team players’ statistics and their recent performance taking into account opponents' strength and tournament level. We assign different weights to all the factors in the model and optimize it to achieve the best predictive results of a win in pairwise comparison.
Using our tool we have programmed possible scenarios of PGL Major Play-ins in the Swiss format. Each team is randomly assigned a win or a loss according to our probabilities and is moved to the next games according to the tournament system, so the opponents of each team are equivalent to what would happen in the real thing. This way, have "played out" the tournament over 200 000 times using our pairwise win probability tool. We have stored the results of each simulation and calculated how many iterations each team is advancing to the main tournament. The share of simulations where a team finishes with 3 wins, is the probability of finishing in the top8 we assign for each team.
The drawbacks of our model are that it assigns the most weight to recent results, which overall is a good thing, but cannot capture, that e.g. some teams might have not taken some matches seriously in preparation for the major. It also does not care for players' experience and their ability to cope with stress, which is an important factor during an event of such rank as the PGL Major. Lastly, as some teams compete only in their rather hermetic regions (e.g. Tyloo) their results may be skewed, as the model considers their opponents as quite weak and has no opportunity to update that in comparison to other regions.
DISCLAIMER! GGPredict is a company that provides AI coaching and analytics for individual players. We are not bookmaking or a forecasting company. These results are in no way betting recommendations. The research was made solely for purpose of our data team's fun (and hopefully yours as well).