Using Data Science to Win CSGO Matches on Mirage
Introduction
I used to be a good player. On a good day I could even have been regarded as one of the best, and I was lucky enough to get to travel the world playing my favorite computer game Counter-Strike. Unfortunately this was in the early 2000s, and winning an international tournament netted you about $1,000. Not exactly something you could make a living off.
Fast forward to 2018, and Goldman Sachs tells us eSports viewership will surpass traditional sports like NFL by 2022. In terms of popularity, longevity and prize pools, few games can match Counter-Strike: Global Offensive. So this seems like a good time to get back into the game and go professional! But with more than 11 million people playing the game last month alone, and professional players competing in tournaments with prize pools as high as $1,000,000, how do you stand out from the crowd?
CSGO is an objective-based first-person shooter, where two opposing teams of Counter-Terrorists (CT) and Terrorists (T) compete to complete objectives like securing a location to plant or defuse a bomb. At the end of each round players are rewarded with in-game currency based on their individual and the teams performance, to purchase more powerful weapons in subsequent rounds.
Professional teams have analysts and coaches to help the players improve. This is luxury you and I can only dream of. But perhaps there are other ways of gaining an advantage over other players? Using a data set of more than 1400 matchmaking games on the popular map de_mirage, I employ data science techniques and use python scripts to attempt to draw insights from the data and answer three questions:
- Which site should you defend?
- Which pistols are preferred in the pistol round?
- Should you buy a defuse kit in the pistol round?
Which site should you defend?
Winning a round in CSGO is achieved by completing one of two objectives: Eliminating all enemy players, or by detonating (Ts) or defusing (CTs) a bomb. Mirage has two bomb sites, A and B, where the Ts can plant the bomb. CTs and Ts start each round in separate parts of the map. Given two sites to choose from, where should the CTs focus their defenses?
By using a python script on a data set of more than 9200 CSGO-rounds played on Mirage, I have plotted heatmaps of player positioning at the time of dealing damage to an opponent. Can this visualization help us in organizing our defense of the bomb sites?
Looking at the CT’s positioning first, there’s hardly any surprises here. Action seems fairly evenly spread between the sites. However, the T-side plot is more interesting, revealing a clear focus towards bomb site B, and B Apartments in particular. Other action areas include T-ramp, Tetris and Palace on site A. Middle, thought to be a vital area to control on Mirage, seems to clearly favor the CT’s superior long-range position in Window, with few successful damage instances from the Ts.
Let’s compare this to a heatmap of players using rifles and SMGs instead, to try and spot any relative differences.
Inspecting the heatmap we can clearly see less emphasis from the Ts on site B, as they turn their efforts towards Middle and site A. It is also interesting to note the heatmap borders expanding when players are equipped with rifles. This is an indication of CTs taking the fight to the Ts, by pushing T-ramp and Middle.
So what effect could this have on our strategy as CTs? A common way to defend on many maps is by having two players on each bomb site, and one player in the middle between the sites. Looking at the data, the natural approach would be to position more CTs towards site B in the pistol round, while focusing more on site A in rounds where players have rifles or SMGs. By overloading the B site with players you might surprise the Ts and win the round!
Which pistols are preferred in the pistol round?
As expected the USP/P2000 and Glock are by far the most common pistols, enabling players to buy Kevlar body armor or utility items like grenades. Somewhat surprising is FiveSeven being used by more than 6 % of the players, while the normally popular CZ is as low as 2 %. This might be due to the data being gathered in a patch where the FiveSeven was relatively more powerful than it currently is.
Should you buy a defuse kit in the pistol rounds?
Given a starting budget of $800, spending $400 on a defuse kit means you will lose out on Kevlar body armor, making you more susceptible to incoming damage. For the defuse kit to have any value in a pistol round, the Ts actually need to plant the bomb. So what does the numbers tell us?
The bomb is planted in 27 % of all pistol rounds, compared to 16 % of normal rounds. This indicates the relative value of having a defuse kit being higher in pistol rounds than other rounds. Even though buying a kit takes up a significant portion of your budget, at least one player on your team should buy a kit.
What about where on the map the defuse kit is most likely to be needed? As we would expect from our analysis above, the bomb is planted on the site B almost 60 % of all bomb plants. Consistent with our previous findings, this picture is turned on its head in normal rifle rounds, where the bomb is more likely to be planted on site A.
Conclusion:
In this post, we took a look at how a data driven approach to CSGO might give us some advantages to make up for old age.
- We looked at player movements in pistol rounds compared to other rounds, and found that site B is a popular target for the Ts in pistol rounds. We also found that in normal rounds site A is more likely to be targeted by the Ts.
- Players tend to prefer the default pistols in the pistol rounds, enabling them to purchase armor or utility.
- The bomb gets planted more frequently in pistol rounds than in normal rounds. Make sure your team has at least one defuse kit! Another finding was that the bomb is more likely to be planted on B in pistol rounds, and on A in subsequent rounds.
Some of these findings are as expected, and some should make us think about how we set up our team for success and the need for a dynamic approach to positioning.
