Steam Game Player Insight

This repository is dedicated to analyzing player behavior and achievement data from Steam games.
The goal is to apply data science, machine learning, and statistical analysis to understand how players engage with different games, discover playstyle archetypes, and study the relationship between playtime, achievements, and retention.

The project is purely personal-interest driven: I started with 7 Days to Die (7DTD) because I’ve recently enjoyed watching it streamed, even though I’ve never played it myself. More games will be added in the future as case studies.

---

Motivation

This repository is a personal research portfolio combining gaming and data science.
Even without playing the games myself, I enjoy exploring how player data tells a story about motivation, retention, and design.
The ultimate goal is to develop reusable frameworks for analyzing game engagement across different genres.

---

Repository Structure

This section provides a detailed overview of the project directory structure, helping you understand where key files and resources are located.

.Steam-GamePlayer-Insight/
├── data/                   # Data storage and processing
│ ├── raw/                  # Original raw data (as collected)
│ ├── intermediate/         # Cleaned / partially processed datasets
│ ├── players/              # Player-level data extracts
│ └── etl/                  # ETL scripts or transformed outputs
│
├── notebooks/              # Jupyter notebooks for analysis
│ ├── 7DTD/                 # Analyses specific to 7 Days to Die
│   ├── 01-EDA.ipynb
│   ├── 02-efficiency-regression.ipynb
│   ├── 03-Player-Segmentation.ipynb
│   ├── 04-PvP-PvE-Classification.ipynb
│   ├── 05-death-playtime-ABtest.ipynb (WIP)
│   └── readme.md
│
├── sql/                    # SQL queries for database extraction or transformation
│
├── LICENSE
├── .gitignore
└── README.md # This file

---

Current Focus: 7 Days to Die (7DTD)

Within notebooks/7DTD/, several studies have been completed:

1. Exploratory Data Analysis (EDA)

   • Cleaned, validated, and explored the dataset.
   • Studied achievement distributions, playtime skew, and data quality issues.

2. Achievement Efficiency Regression

   • Modeled efficiency score (achievements per playtime).
   • Polynomial regression (degree 3) achieved R² ≈ 0.96.
   • Random Forest Regression confirmed log_playtime, crafting_count, combat_count as key predictors.
   • Showed potential use for anomaly/cheater detection.

3. Player Segmentation (Clustering)

   • PCA (81% variance explained) + hierarchical clustering → 6 archetypes:
     • Generalists, Crafters, Explorers, Fighters, Death-heavy quitters, Steady-progress.
   • Identified high-churn groups (death-heavy) and content-focused niches (crafters/explorers).
   • Provided insights for retention strategies and design balance.

4. PvE vs PvP Classification

   • Defined groups via player kill data.
   • A/B testing showed multi-player mode players have significantly higher playtime.
   • Classification models: Logistic Regression (ROC-AUC 0.90), Random Forest (ROC-AUC 0.95).
   • Highlights the importance of multiplayer for retention.

5. Death vs Playtime A/B Testing (Work-in-progress)

   • Investigating whether early deaths (in the first N unlocked achievements) impact long-term playtime.
   • Aim: detect frustrated quitters at an early stage.

---

Future Directions

• More Games: Extend the framework to additional Steam titles (MMOs, PvP shooters, sandbox survival).
• Sequential Modeling: Use RNNs (LSTM/GRU) to predict achievement unlock sequences, progression paths, and churn.
• Cheater/Bot Detection: Leverage efficiency scores and outlier behavior to flag abnormal players.
• Cross-Game Archetypes: Compare whether playstyle clusters generalize across different genres.

---