🧠 White Box - Supervised ML Project

📘 Overview

This project demonstrates the development of two types of supervised machine learning models — Regression and Classification — using real-world datasets.
Each model follows a structured workflow: Exploratory Data Analysis (EDA) → Preprocessing → Model Building → Evaluation → Interpretation.

---

🎯 Regression Model — Video Game Sales Prediction

Objective

Predict global video game sales based on regional sales data (North America, Europe, Japan, and Others) using regression techniques.

Models Used

1. Linear Regression

   • Version 1: Used 3 numerical features (NA_Sales, EU_Sales, JP_Sales).
   • Version 2: Added Other_Sales and Year to improve predictive accuracy.
   • Evaluation Metrics: R² Score, RMSE, and Residual Analysis.
   • Outcome: The model performed well in explaining the relationship between regional and global sales, showing strong correlation across regions.

2. K-Nearest Neighbors (KNN Regressor)

   • Tested with k=3 and k=5 neighbors.
   • Performance compared using R² and RMSE.
   • While the model handled non-linear patterns, it was sensitive to scaling and dataset size.
   • Best result: Linear Regression (5 features) showed the highest accuracy and interpretability.

---

🧾 Classification Model — Credit Score Prediction

Objective

Predict a customer’s Credit Score category (Good, Standard, or Poor) based on financial and demographic data.
The model helps financial institutions automate risk assessment and reduce human bias.

Models Used

1. Logistic Regression

   • Version 1: Used only numeric features.
   • Version 2: Included both numeric and encoded categorical features.
   • Metrics: Accuracy, Precision, Recall, F1 Score.
   • Result: Performed best overall — interpretable and consistent across all metrics.

2. K-Nearest Neighbors (KNN Classifier)

   • Version 1: k=3
   • Version 2: k=5
   • Evaluated with the same metrics.
   • KNN showed good performance after scaling but was less stable compared to Logistic Regression.

---

📊 Evaluation Summary

Model Type	Version	Accuracy / R²	F1 / RMSE	Remarks
Linear Regression	3 features	High R²	Low RMSE	Simple & effective
Linear Regression	5 features	Best R²	Lowest RMSE	Most reliable
KNN Regressor	k=3 / k=5	Moderate	Medium RMSE	Non-linear, less stable
Logistic Regression	Numeric only	Balanced	Good F1	Solid baseline
Logistic Regression	All features	Best overall	Highest F1	Recommended
KNN Classifier	k=3 / k=5	Moderate	Fair F1	Sensitive to scaling

---

🧩 Tools & Libraries

• Python, Pandas, NumPy, Matplotlib, Seaborn
• scikit-learn for model training and evaluation
• Jupyter Notebook for code, visualizations, and documentation

---

🧠 Key Takeaways

• Linear and Logistic Regression models are interpretable, efficient, and perform consistently well.
• KNN can capture non-linear relationships but requires careful scaling and parameter tuning.
• EDA and preprocessing are essential steps to ensure model reliability.