Global Mobile Gaming: End-to-End Data Analysis & Strategic Marketing

This project simulates a Data Analytics Consultant role for a global mobile gaming company. It covers the full analytical lifecycle — from raw data manipulation and realism engineering, through exploratory analysis, to machine learning and a targeted marketing strategy.

Developed as a Final Project for the Miuul Data Analytics Bootcamp.

📌 Project Overview

Key work:

• Data Manipulation & Realism Engineering: The raw Kaggle dataset had unrealistic distributions (e.g. equal gender split in every region). Purchase rates, spend amounts, region distribution and genre mix were re-generated with realistic, seeded (random_state=42) rules to make the data analysis-ready.
• Exploratory Data Analysis (EDA): Investigated player behavior, regional differences and in-game spending with pivots, cross-tabs and visualizations.
• Segmentation & Personas: K-Means segmentation of players into behavioral personas.
• Machine Learning: Classification models with an explicit data-leakage analysis (see below).
• Visualization & Reporting: A 3-page interactive Power BI dashboard for executive decision-making.

🤖 Machine Learning — Honest Results

The ML section is split into three parts, all running on a single consistent dataframe.

1. Player Segmentation (K-Means)

Clustering on scaled numerical behavioral features only (categoricals are used for profiling, not distance). Silhouette scores are low (~0.12) across all k, showing the (largely synthetic) data has no sharp natural clusters — reported transparently. k = 3 is used to produce interpretable personas: Casual, Regular, Hardcore.

2. Engagement Classification — a leakage case study

EngagementLevel is not an independent label; it is rule-derived from session features (sessions/week, session duration, playtime). Predicting it from those same features yields ~92% accuracy — but this is label reconstruction, not prediction. We prove it:

Feature set	Accuracy
All features (incl. session features)	~0.925
Session features removed	~0.595

The 92% collapses to 60% once the derived-from features are dropped → data leakage confirmed. This label is best used for reporting/segmentation, not as a prediction target.

3. In-Game Purchase Prediction — the real model ✅

A leakage-free, business-relevant target: will the player make an in-game purchase? (InGamePurchases, ~20% positive).

• ROC-AUC: ~0.82
• PR-AUC: ~0.68 (base rate 0.20)
• Top drivers: Location and GameGenre

Business takeaway: marketing budget should target high-propensity region × genre combinations (e.g. Asia/Europe + Strategy) to maximize ROI.

📊 Power BI Dashboard

A 3-page interactive dashboard:

1. Global Overview — high-level KPI tracking
2. Behavioral Analysis — session patterns and engagement
3. Regional Spending — geographic revenue distribution and genre preferences

🛠 Tech Stack

• Language: Python (Pandas, NumPy, SciPy, Scikit-Learn)
• Visualization: Power BI, Seaborn, Matplotlib
• Workflow: Scikit-Learn Pipeline + ColumnTransformer (imputation, scaling, one-hot encoding)

📁 Repository Structure

├── data/                 # Raw and processed datasets
├── notebooks/            # EDA and machine learning workflow
├── reports/              # Power BI dashboard (.pbix) and screenshot
├── images/               # Supporting visuals
├── requirements.txt      # Project dependencies
└── README.md             # Project documentation

▶️ Getting Started

pip install -r requirements.txt
jupyter lab notebooks/gaming_behavior_analysis.ipynb