World of Warcraft

1. Objective Predict Game Player Churn

Classify game player who is likely to be churned in six month

The experiment aims to prevent game players from quitting within a month by offering them a discounted 6-month subscription.

Predicting game player churn is an important problem in the gaming industry as it helps game developers and publishers understand the behavior of their players and take necessary steps to retain them. By analyzing the pattern of players before they churn, game developers can take proactive measures to retain their players and prevent them from quitting the game.

2. Data source

Kaggle (World of Warcraft Avatar History | Kaggle)

The techniques you expect to use in your analysis

Use Classification estimator and conduct hyperparameters optimization.
Operationalize the model as Rest API

3. Exploratory Data Analysis

Details EDA Notebook

Data has 27,681 game players information

Game player's character level

Level 70 was maximum cap before November 2008 which explain why there are many activity of level 70

Races in game

Blood Elf is the most popular race in the game

Level by Race

In November, the maxium level was increase to 80, many palyers back to play to reach the level to 80
Except Blood Elf, most of races show similar trend

Character Classes

Guild

-1 indicate no guild, never join any guild
Guild 103 is the most popular and 282 is second

4. Business Findings

Number of guild activity and game play

Low level gamers move guild often than high level gamers
High level gamers change guild less than 5 times a year

Subscription model

World of Warcraft (WoW) offers 3 diffenent subscriptions

- 1 month, $ 14.99/month
- 3 months, $ 13.99/month
- 6 months, $ 12.99/month

Free trial up to level 20

Period of playing game in months

55.2 % of player drop after a month of game play
When the gamer pass 8 months, they might play to 4 or more
13-months mean that the gamer played the game since January to December, the first month counts a 1

Number of players by period of playing by race

New players

Every month there are more than 1,000 of new players (or new characters)
In October there was new update and it caused new character Orc and Troll races

Level activity

When the gamer reaches level 60+, number of zone for game play reduced

Game play and level

Game players who played the game 12 months or more show more activity for both game play and guild

Player level up journey

5. Training

Details Experiment Notebook

Classification Baseline models

Estimator	training(sec)	Train Score	Test Score	Precision	Recall
Logistic Regression	0.07	0.68	0.68	0.61	0.65
Decision Tree	0.04	0.99	0.66	0.59	0.61
KNeighbors	0.01	0.77	0.66	0.60	0.56
SVC	1.39	0.63	0.63	0.54	0.85
RandomForest	0.57	0.99	0.71	0.66	0.64

RoC Curve

RandomFroest shows the best result among the estimators

Hyperparameter Tuning

Logistic Regression

Best hyperparameters: {'estimator__C': 10, 'estimator__penalty': 'l1', 'estimator__solver': 'liblinear'}
Test set accuracy: 0.72

RandomFroest

Best hyperparameters: {'estimator__max_depth': 7, 'estimator__max_features': 'sqrt', 'estimator__min_samples_leaf': 3, 'estimator__min_samples_split': 2, 'estimator__n_estimators': 50}
Test set accuracy: 0.7397525135344161

Next step

Action items

Use regression estimator to predict game play preiod

Use ML Interprt to understand festure importance

xlegend1024/wow-player-churn