Who are the top ESL Impact players' doppelgängers?
We looked into some of the best women in the game and found their doppelgängers ahead of the third season of ESL Impact.
The ESL Impact circuit kicked off in 2022, giving opportunities for the best women's Counter-Strike players to compete against each other on LAN. Just a few tournaments in, we have enough data to see who the best players are, as well as the style of those players: Are they aggressive? How much do they AWP? Do they have a knack for the clutch?
Ahead of ESL Impact Season 3, which starts this week, we are going to try and answer those questions. Rather than answering with the raw figures, however, we will do something a bit more fun and pair each ESL Impact player with a male statistical doppelgänger.
We use statistics as basic as kills per round, as well as things like grenade damage or the team's win percentage after they get a kill, to build a full picture of what type of a player someone is — and then we can find their closest match.
This should show you, even if you have not watched a single ESL Impact match, at least a little bit about how the women's scene's best players like to play Counter-Strike.
When we did this for the WePlay Academy League we used a basic match system, assigning points for each 'match' a player had to another in each metric. This time we are using a more complicated but more robust mathematical method, namely cosine similarity, that can look at each player in their entirety.
We are also adding a new first step, a cluster graph of every player in our sample. This is a compilation of every stat we used in our comparison, presented in 2D. There is not a 'good' or 'bad' corner; the graph simply groups players that are similar to each other.
As you can see, the algorithm has done a pretty good job with basic divides. AWPers are in the same group in the top right, with a less clear-cut divide between star and supportive riflers to make up the remaining clusters.
The reason we include this is to provide some context. Our doppelgängers are the most similar player available to the chosen target. They are not always a 100% match, especially when we are analyzing unique players like Olga "Olga" Rodrigues who are both super-aggressive and super-consistent.
The same is true of Dzhami "Jame" Ali, who is also in an island of his own but this time at the top of the chart, a polar opposite to Olga and Mareks "YEKINDAR" Gaļinskis down south.
Ana "ANa" Dumbravă
The reigning women's player of the year, ANa, is paired with fellow superstar AWPer Dmitry "sh1ro" Sokolov. Both players are exceptional with the big green, and also use it quite often; ANa boasted a 0.51 AWP KPR in our sample and a survival rate of 50% which puts her closer to sh1ro than to the rifle-happy Oleksandr "s1mple" Kostyliev and Mathieu "ZywOo" Herbaut.
Both ANa and sh1ro go for relatively few opening duels (17% and 16%) but come out with huge success rates — ANa 76% and sh1ro 69%. They are efficient, low-risk, statistical phenoms with proficiency in nearly every aspect of the game.
ANa's five most similar players:
Dmitry "sh1ro" Sokolov: 96.8%
Helvijs "broky" Saukants: 95.8%
Ilya "m0NESY" Osipov: 94.8%
Mathieu "ZywOo" Herbaut: 94.6%
Abdul "degster" Gasanov: 94.0%
Olga "Olga" Rodrigues
As we said in the introduction, Olga is a unique player. Throughout 2022 she averaged a 1.39 rating, 1.60 Impact, 99.7 ADR, and 0.86 KPR. These are not numbers you see in the tier one circuit, even for the game's very best AWPers like s1mple and ZywOo.
What makes this even more impressive is her aggression, averaging 29.7% opening kill attempts and 0.17 opening kills per round. Her status as an aggressive yet ultra-consistent rifler sees our algorithm link her to the consistent stars in Nikola "NiKo" Kovač and Sergey "Ax1Le" Rykhtorov just slightly more than the aggressive YEKINDAR (96.0%) and Robert "Patsi" Isyanov.
Olga's five most similar players:
Nikola "NiKo" Kovač: 98.2%
Sergey "Ax1Le" Rykhtorov: 97.2%
Jonathan "EliGE" Jablonowski: 97.0%
Håvard "rain" Nygaard: 96.9%
Abay "HObbit" Khassenov: 96.8%
Ksenia "vilga" Kluenkova
vilga is most similar to the other Kovač, Nemanja "huNter-" Kovač. They both have strong impact in wins, sharing the same kills per won round (0.98), and are also nearly identical when it comes to assisted kills (18-19%).
vilga is more aggressive than huNter-, with 6% more opening kill attempts and surviving for 12 fewer seconds a round on average; our algorithm has paired them thanks to vilga's consistency making her statistical profile resemble a more passive player. 35% survival and 77% KAST are two such metrics, rarely high figures for a team's most aggressive T side player.
vilga's five most similar players:
Nemanja "huNter-" Kovač: 96.2%
Paweł "dycha" Dycha: 96.1%
René "TeSeS" Madsen: 95.9%
Martin "stavn" Lund: 95.9%
Sergey "Ax1Le" Rykhtorov: 95.8%
Mounira "GooseBreeder" Dobie
GooseBreeder is paired with three Heroic players in her top five, probably due to her impressive trade stats. She recorded 0.14 trade kills per round in 2022, as well as 0.15 traded deaths, as well as having 33% of her opening deaths be traded. This clearly identifies her as a pack player, with good spacing and constantly in and around the action.
This is not unusual for an IGL like GooseBreeder, and neither is her strong flashbang statistics: She blinded opponents for 2.59 seconds per round, with 0.09 flash assists. This is another metric that links her to Heroic, where every single rifler contributes in both trades and utility.
GooseBreeder's five most similar players:
Martin "stavn" Lund: 96.3%
Rasmus "sjuush" Beck: 96.3%
Yuri "yuurih" Santos: 96.2%
René "TeSeS" Madsen: 96.0%
Denis "electroNic" Sharipov: 95.6%
Angelika "Angelka" Kozłowska
Angelka of NAVI Javelins has erupted into form in 2023, averaging a 1.31 rating at ESL Impact Katowice. For 2022 however, her closest match was EliGE. Both sported strong ADR, KPR, and Impact while lagging behind in DPR and KAST last year, hinting at their status as fairly aggressive riflers.
They were not super-aggressive, though, both operating as the second or third most aggressive (by opening attempts) riflers in their team. They are glue riflers, flexible in role but consistent in output.
It is true that EliGE's role changed a fair bit during 2022, acting as Liquid's most aggressive rifler in the Richard "shox" Papillon era. This, when combined with Pavle "Maden" Bošković and Kristian "k0nfig" Wienecke's presence in the top five, illustrates how Angelka tilts slightly towards the aggressive side of things when it comes to riflers.
Angelka's five most similar players:
Jonathan "EliGE" Jablonowski: 97.7%
Martin "stavn" Lund: 97.2%
Paweł "dycha" Dycha: 97.0%
Pavle "Maden" Bošković: 97.0%
Kristian "k0nfig" Wienecke: 96.9%
You can find every other player's doppelgänger here. The sheet also shows a similarity matrix for each player, showing a full list of the most similar men for each woman.
First, we built a database of players, men and women, using 2022 stats, with 52 metrics including but not limited to:
- Assisted Kill%
- Deaths per round
- Time alive per round
- AWP kills per round
- Damage per round
- Headshot percentage
- Impact rating
- Opening kills (T and CT)
- Opening attempts (T and CT)
- Trade kills
- Trade deaths
- Opponents flashed duration
- Grenade damage
- Clutch points
- Win% after a kill
Then we used K-means to create cluster groups, the groups that are visualised in 2D, using PCA with the scatterplot at the start of the article.
The similarity scores used to identify doppelgängers are calculated differently, using cosine similarity to compare players over lots of metrics. This was calculated differently for riflers and AWPers, with slight changes to the statistics we used.
The limiting factor to this process' usefulness are the statistics available. Even with 52 different metrics we cannot capture what makes a player unique. We also cannot divorce a player from their team; women that performed well in trading statistics were more likely to match a Heroic player, even if they had different roles in game, for example.
All of our similarity scores were quite high, with the lowest match still being in the high 70s. This could be due to a number of factors, but ultimately cosine similarity still produced better results than euclidean distance or our original 'match' method. Thus, we left our 'similarity score' as the raw cosine distance.