💤 Forecasting Sleep Efficiency using Machine Learning

COMP 542 Project
A machine learning-driven analysis of behavioral, physiological, and lifestyle factors to predict sleep efficiency.

---

🧑‍🤝‍🧑 Team Members

• Brandon Ismalej
• Jittapatana (Patrick) Prayoonpruk
• John Lee

---

🧠 Project Overview

Sleep efficiency is a vital indicator of sleep quality and overall health. This project utilizes advanced machine learning algorithms to forecast sleep efficiency using behavioral and lifestyle-related data. It aims to assist individuals and healthcare professionals in identifying patterns that influence sleep and promote better sleep hygiene.

---

📦 Dataset

• Source: Kaggle - Sleep Efficiency Dataset
• Samples: 452 observations
• Features: 14
  • Sleep duration, REM %, deep/light sleep %, caffeine/alcohol consumption, smoking, gender, awakenings, etc.
• Target: Sleep Efficiency (continuous)
• Train-Test Split: 80% train / 20% test
• Cross Validation: k-fold with k=5

---

🧹 Data Preprocessing

• Categorical Encoding: Binary encoding for gender and smoking status
• Missing Values: Imputed using mean/median values within groups
• Feature Selection Techniques:
  • Correlation Analysis
  • Mutual Information
  • Recursive Feature Elimination (RFE)
  • Random Forest Importance

---

🎯 Selected Key Features

Feature	Correlation	Impact Description
Light Sleep %	-0.82	Strong negative impact on sleep efficiency
Deep Sleep %	+0.79	Strong positive correlation (most restorative)
Awakenings	-0.55	More awakenings reduce sleep quality
Smoking Status	-0.29	Behavioral factor negatively affecting sleep

---

🔍 Models Used

1. Multiple Linear Regression (with stepAIC)

• Why: Simplicity, interpretability, and effectiveness for linear patterns
• Metric: Adjusted R², AIC, BIC
• RMSE: 0.0617

2. XGBoost (Gradient Boosting)

• Untuned RMSE: 0.0563
• Tuned RMSE: 0.0494 (best performing model)

3. LightGBM

• All Features RMSE: 0.0511
• After Feature Selection RMSE: 0.0525

4. Random Forest

• All Features RMSE: 0.0503
• With Feature Selection RMSE: 0.0557

---

🧪 Model Evaluation

Model	RMSE	Notes
Multiple Regression	0.0617	Stepwise selected features
XGBoost (untuned)	0.0563	Baseline GBM
XGBoost (tuned)	0.0494	⭐️ Best performance
LightGBM (all features)	0.0511	Fast and efficient
Random Forest	0.0503	Robust ensemble model

---

📈 Tools & Libraries

• Python: scikit-learn, xgboost, lightgbm, pandas, matplotlib
• R: stepAIC, ggplot2, lm(), diagnostic plotting
• Jupyter & RMarkdown for exploratory and final analysis

---

🚀 Potential Applications

• Personalized sleep tracking and health monitoring
• Integration into smartwatches, fitness trackers, or health apps
• Use in sleep disorder clinics for patient screening

---

🧩 Future Enhancements

• Deploy models in a web/mobile interface
• Integrate additional physiological signals (heart rate, HRV)
• Use time-series modeling for longitudinal sleep data

---