Projet-3: Video Game Sales Prediction with Linear Regression 📊

📌 Project Overview

Can we predict the global sales of a video game based on its product features and player engagement metrics? This project combines two Kaggle datasets:

• Steam game data (price, playtime, reviews, scores…)
• Global video game sales data

The objective is to build a supervised machine learning model capable of estimating Global_Sales using measurable indicators available at or shortly after release.

🎯 Objectives

Merge Steam and global sales datasets Identify key variables influencing commercial performance Build a predictive model (Linear Regression) Evaluate performance using MSE & R² Extract actionable business insights

📊 Dataset & Features

🎯 Target Variable Global_Sales

📌 Explanatory Variables Price Required_Age Achievements Average_Playtime Median_Playtime User_Score Critic_Score Positive_Reviews Negative_Reviews

These variables capture both:

• 🎮 Product attributes
• ⭐ Quality perception
• ⏱️ Player engagement
• 🔍 Exploratory Data Analysis (EDA)

📈 TL;DR

🔎 Quick Exploration (EDA)

- Strongest correlation with Global Sales: Positive_Reviews and Critic_Score show the most significant positive relationship with commercial performance.

- Engagement is more predictive than price: Average_Playtime demonstrates a meaningful link with sales, suggesting that retention depth reflects economic value.

- Price impact remains limited: Price alone does not strongly explain performance. High-priced games do not systematically outperform lower-priced titles.

- Review volume acts as a visibility multiplier: The total number of reviews (positive + negative) is strongly associated with sales, indicating an exposure dynamic beyond pure sentiment.

📊 Business Interpretation

- Scores vs Sales: Higher User_Score and Critic_Score tend to align with stronger commercial performance, but quality perception alone is insufficient.

- Engagement depth matters: Titles with longer median and average playtime generally generate stronger global revenue.

- Exposure dynamics: Review volume likely amplifies algorithmic visibility, reinforcing a feedback loop between player engagement and sales.

- Structural limitation: The model captures measurable engagement and perception signals, but does not integrate: Marketing intensity, Franchise power, Platform exclusivity, Release timing

🤖 Model

Model type: Linear Regression (Scikit-Learn) Train/Test split Performance metrics: MSE (Mean Squared Error) / R² Score

The model explains a significant portion of variance in Global_Sales, while highlighting the limits of structured quantitative data in entertainment industries.

💡 Business Insights

• Engagement metrics are strong commercial indicators.
• Volume of reviews acts as a visibility amplifier.
• Quality perception influences sales, but is not sufficient alone.
• Predictive models should combine behavioral + contextual features.

👉 Commercial success is a mix of engagement + perception + exposure.

📈 Visualizations

🚀 Deliverables

• 📓 Full Jupyter Notebook
• 📊 Data visualizations
• 🤖 Predictive model
• 📝 Executive summary (this README)

📂 Data Source

All data used comes from Kaggle :

👉 Link to the dataset "Steam's daily sales"

_{All rights belong to their respective authors.}