😴 Sleep Health Statistics & ML Prediction

Statistical analysis and Machine Learning prediction of sleep quality based on lifestyle and health data.

---

Project Overview

This project combines statistical analysis and Machine Learning to study and predict sleep quality from a dataset of 374 individuals.

It follows a structured methodology inspired by CRISP-DM and covers the full data science pipeline — from descriptive statistics to predictive modeling.

---

Objectives

• Explore and understand sleep health data
• Perform descriptive statistical analysis (mean, variance, confidence intervals)
• Apply hypothesis testing
• Build and evaluate ML classification models to predict sleep quality
• Interpret results and extract meaningful insights

---

📂 Project Structure

sleep-health-statistics/
│
├── data/                   # Raw dataset + extracted columns
├── notebooks/              # Jupyter notebooks (stats + ML)
├── machine_learning/       # ML models and results
├── requirements.txt        # Dependencies
└── README.md

---

Part 1 — Statistical Analysis

Variable studied: Sleep Duration (hours/day)

Statistic	Value
Min	5.8h
Max	8.5h
Mean	7.13h
Variance	0.63
Std Dev	0.80h
CI 95%	[7.05 , 7.21]
CI 98%	[7.03 , 7.23]

Hypothesis Test (t-test):

• H0: mean = 7h → Rejected (p = 0.0014 < 0.05)
• Conclusion: individuals sleep significantly more than 7h on average

---

Part 2 — Machine Learning

Problem

Can we predict sleep quality from lifestyle and physiological features?

Type: Multi-class classification (Poor / Average / Good)

Pipeline

1. Data Cleaning — handling missing values in Sleep Disorder
2. Feature Engineering — creating Sleep_Quality_Category from score
3. Outlier Detection — IQR method (15 outliers in Heart Rate, kept)
4. Encoding — One-Hot Encoding for categorical variables
5. Normalization — StandardScaler on numerical features
6. Train/Test Split — 80% train / 20% test (stratified)

Model — Logistic Regression

Metric	Score
Accuracy	98.67%
Precision (avg)	0.99
Recall (avg)	0.99
F1-score (avg)	0.99

Key findings:

• Average and Good classes predicted with near-perfect accuracy
• Poor class correctly identified despite being underrepresented
• Very few misclassifications confirmed by the confusion matrix

---

Technologies

---

Author

BENZHIR Wafa — Bachelor's student in Data Science & Intelligent Systems

---

Academic Context

Project completed as part of the Statistiques Avancées module 🎓 Academic year: 2025–2026