Game Player Churn Prediction Project Overview

This project builds a machine learning model to predict player churn using simulated game telemetry data. The goal is to identify behavioral signals that indicate retention risk and generate actionable product insights.

The dataset simulates 5,000 players with engagement, progression, monetization, and recency features.

🧠 Problem Statement

Player churn significantly impacts revenue and long-term growth in games. The objective is to: Predict which players are likely to churn Identify key drivers of churn Propose retention strategies based on data insights 📊 Dataset Features

Engagement

days_active_last_30

total_sessions

avg_session_minutes

Progression

max_level_reached

total_upgrades

Monetization

total_spend

Recency

last_login_days_ago

Categorical

country

device_type

Target variable:

churned (0 = active, 1 = churned)

⚙️ Modeling Approach

Label Encoding for categorical variables

Train/Test split (80/20)

Random Forest Classifier

Evaluation using:

Accuracy

Precision / Recall / F1

ROC AUC

Confusion Matrix

Cross-validation for stability

📈 Model Performance

ROC AUC: 0.91

Accuracy: 0.82

Cross-validated ROC AUC: ~0.89–0.91

The model demonstrates strong predictive capability without feature leakage.

🔍 Key Findings

Top churn drivers:

Total sessions

Last login recency

Total spend

Days active last 30

Players with low engagement frequency, high inactivity, and low monetization activity show the highest churn probability.

🎯 Product Recommendations

Trigger re-engagement campaigns after 5–7 days of inactivity

Provide progression boosts to low-session players

Offer retention-focused promotions to mid-spend cohorts

Monitor engagement drop-offs early in lifecycle

🚀 Future Improvements

Compare with XGBoost / Gradient Boosting

Add SHAP explainability

Deploy as a Streamlit dashboard

Extend to Lifetime Value (LTV) prediction

🛠 Tech Stack

Python

Pandas

NumPy

Scikit-learn

Matplotlib

Seaborn