🎮 Game Player Churn Prediction

A full end-to-end machine learning project that predicts whether a game player is likely to churn (stop playing) based on their in-game behavior.

Problem Statement

Gaming companies lose revenue when players stop engaging. This project builds a churn prediction system that identifies at-risk players early so targeted retention strategies can be applied.

Dataset

• Source: Predict Online Gaming Behavior Dataset
• Size: 40,034 players × 13 features
• Target: EngagementLevel (Low / Medium / High)

What's Built

• Exploratory Data Analysis (EDA) with visualizations

• Feature Engineering 11 new features created from raw data

• New data size 40034 * 30

• 4 Models trained and compared: Logistic Regression, KNN, Random Forest, XGBoost | Model | Accuracy | AUC |

  | XGBoost | 92% | 0.946 | | Random Forest | 89.4% | 0.921 | | KNN | 82.1% | 0.891 | | Logistic Regression | 78.3% | 0.845 |

• Full evaluation: Confusion Matrix, Classification Report, ROC/AUC

• Interactive Streamlit dashboard for live predictions

Key Results

• 92% accuracy on unseen test data
• AUC 0.946 for churn class detection
• SessionsPerWeek and AvgSessionDurationMinutes are the strongest churn predictors (72% of model decisions)
• 6 of top 15 features were engineered — validating domain-driven feature creation

Tech Stack

Python · XGBoost · Scikit-learn · Streamlit · Pandas · Matplotlib · Seaborn