Kaggle released a dataset of over 65,000 games' worth of anonymized data points consisting of number of kills, distance traveled and so on, along with the victory ranking for the match. The question was: What's the best strategy to win in PUGB?
PUGB is a battle royale game where many players (up to 100) are parachuted onto an island to fight or hide until one (or one team of up to four players) is the last standing. To keep the game interesting, a shrinking bubble is routinely placed on the map so that those outside it will die, forcing players to face each other.
Battle royale games are (a) AWESOME for a quick fix and (b) massively popular outside of just, you know, me :)! PUGB, Fortnite, Apex Legends are just a few to name. Millions and millions of dollars have been spent on battle royale games. But don't just take it from me, check out some spending analysis here: https://www.statista.com/statistics/1078797/battle-royale-in-game-spending-usa-by-game/. In 2020 alone, players spent $ 1.6billion on PUGB and $293 million on Fortnite. Further, almost 350 million people in May 2020 were registered users (https://www.statista.com/statistics/746230/fortnite-players/) of Fortnite worldwide -- that's about the population of the United States, y'all!
Below we have a table of the raw data acquired. A string is simply a sequence of characters, and here they are ID numbers for players, groups, and matches. The type int64 is a signed (+/-) integer type allowing values of -263 to 263-1. float64s work similarly for decimal numbers.
There are no missing entries; instead, there are duplicate entries of players, each anonymized, so that some teams are erroneously shown to be larger than 4. Duplication occurred when a player disconnected (d/ced) or otherwise left the battle and rejoined, so in the aggregate for each match and team the data should be sound.
| name | type |
|---|---|
| Id | string |
| groupId | string |
| matchId | string |
| assists | int64 |
| boosts | int64 |
| damageDealt | float64 |
| DBNOs | int64 |
| headshotKills | int64 |
| heals | int64 |
| killPlace | int64 |
| killPoints | int64 |
| kills | int64 |
| killStreaks | int64 |
| longestKill | float64 |
| matchDuration | int64 |
| matchType | string |
| maxPlace | int64 |
| numGroups | int64 |
| rankPoints | int64 |
| revives | int64 |
| rideDistance | float64 |
| roadKills | int64 |
| swimDistance | float64 |
| teamKills | int64 |
| vehicleDestroys | int64 |
| walkDistance | float64 |
| weaponsAcquired | int64 |
| winPoints | int64 |
| winPlacePerc | float64 |
The data provides some nice raw ingredients for further analysis. Some of them may be useful as directly as features, but we will also derive some. For example:
- headShotRate = headShotKills/kills
- weaponsPerDistance = weaponsAcquired/(walkDistance + rideDistance + swimDistance)
- killsPerWeapon = weaponsAcquired/kills etc.
Another thing to note is the match types. There are campaigns of squads (teams of up to 4), duos (teams of 2), and solo (single-player). Furthermore, there are first-person shooter (fpp) games versus third-person shooter (tpp) games. We'd like to see how each of these game styles, the campaigns and the perspective (which we call pp, person persective), affect the gameplay of individuals and teams.
One shortcoming of the data is that players are anonymized at each entry, so we cannot track a player to compare how the person does in squads versus solo or tpp versus fpp.