Warzone Package: Part 3

3 min readAug 31, 2021

What features have the most robust relationships to team performance?

Let's start by defining team performance. In its simplest state, this can be approximated by final team placement. For our analysis, we will be using the final placement percent as our response variable. This is a value between 0 and 1. With 1 being the team won the match and 0 signifying the team was first to be eliminated. We do this due to the number of teams in each lobby fluctuating. We will be examining Rebirth Quads.

Get Peter Rigali’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

To start we filter our data using the DocumentFilter Class. We what all the data so we use original_df=cod.whole. We set map_choice=’mp_e’ and mode_choice=’quad’ to return only Rebirth Quad matches. This leaves us with a sample size of 50,300 players from 1250 matches. Which averages to near 40 players and 10 teams per match

doc = DocumentFilter(original_df=cod.whole, map_choice='mp_e', mode_choice='quad')

Then we will check the Correlation. For this we will be dropping categorical data types, focusing only on float data type variables.

Below is a Table Plot of the features and respective correlation to placement percent.

Correlation values related to Placement Percent

At the top of our list we TeamSurvialTime with a 0.89. This makes intuitive sense, the longer a team is alive the more likely they will win.

Note noramlize_x = [‘teamSurvialTime’], this normalizes the values between 0 and 1.

Notice the slope value in the legend is 0.95.

As we work our way down the Correlation table, we slowly get away from features that would be affected by the length of the game (timePlayed or distanceTraveled) or based on winning the game (scoreXp or totalXp). Im more interested in things like kills, longestStreak, objectiveBrMissionPickupTablet, or scorePerMinute. Using a similar Scatter to the one noted above, we get.