🏋️ Fitness Classification Project

📋 Problem Description

This project addresses a Binary classification problem using a synthetic dataset that simulates real-world fitness assessment. The goal is to predict whether a person is fit (is_fit = 1) or not fit (is_fit = 0) based on various health and lifestyle features.

📊 Dataset Overview

• 📈 Samples: 2,000 individuals
• 🎯 Features: 10 predictive features + 1 target variable
• ⚖️ Target Distribution: ~60% not fit (0), ~40% fit (1)
• 🚨 Data Challenges:
  • 🔀 Mixed data types (numerical and categorical)
  • ❓ Missing values (~8% in sleep_hours)
  • 🔠 Inconsistent formatting (mixed numeric/string values)
  • 📊 Outliers present in weight_kg (~2% of samples)
  • 🔗 Complex feature relationships with target

🗂️ Features

Feature	Description	Type
age	Age in years	Integer
height_cm	Height in centimeters	Integer
weight_kg	Weight in kilograms (contains outliers)	Integer
heart_rate	Resting heart rate in BPM	Float
blood_pressure	Systolic blood pressure in mmHg	Float
sleep_hours	Average daily sleep hours (contains NaNs)	Float
nutrition_quality	Daily nutrition score 0-10	Float
activity_index	Physical activity level 1-5	Float
smokes	Smoking status (mixed types)	Mixed
gender	Gender ('M' or 'F')	String
is_fit	Target variable (1=fit, 0=not fit)	Binary

🔧 Methodology

🧹 Data Preprocessing

1. 🔄 Handled Mixed Data Types:

   • ✅ Converted 'smokes' column to consistent binary format (0/1)
   • ✅ Created 'is_male' feature from gender column (0/1 encoding)

2. 🎯 Missing Value Treatment:

   • ✅ Imputed missing 'sleep_hours' values with median from training set to avoid Data Leakage

3. 📏 Feature Scaling:

   • ✅ Standardized numerical features for logistic regression model, Tree based models do not require scaling - hence used a copy of same dataset.

4. 🔍 Exploratory Data Analysis:

   • ✅ Generated histograms and boxplots to understand data distribution
   • ✅ Analyzed feature relationships and outlier patterns

🤖 Model Development

Four classification models were trained and evaluated:

Model	Type	Purpose
Logistic Regression	Linear	Baseline performance
Random Forest	Ensemble	Handle non-linearity
XGBoost	Gradient Boosting	State-of-art performance
Decision Tree	Tree-based	Interpretability

📊 Model Performance

🏆 Validation ROC-AUC Scores

Model	🥇 ROC-AUC Score	Rank	Status
Logistic Regression	0.8566	1	🥇 Best Model
Random Forest	0.8417	2	🥈
XGBoost	0.8400	3	🥉
Decision Tree	0.7999	4

⚙️ Hyperparameter Tuning

• 🔧 Performed hyperparameter optimization on training set
• 📋 Validated performance on separate validation set
• 🎯 Logistic Regression demonstrated superior performance

🚀 Deployment

🌐 Production Setup

• 🎯 Final Model: Logistic Regression trained on full training dataset
• ☁️ Deployment Platform: Fly.io
• 🖥️ Application Type: Binary classification web service
• 🎥 Project Deployment Demo: Includes demonstration video

💡 Key Findings

• ✅ Logistic regression achieved the best performance despite dataset complexity
• ✅ Proper data preprocessing was crucial for model success
• ✅ Feature scaling significantly improved linear model performance
• ✅ The synthetic dataset effectively mimics real-world data challenges

FastAPI Prediction Service

A machine learning prediction service built with FastAPI, UV, and Docker.

🚀 Quick Start

# 1. Build Docker image
docker build -t fitness-prediction .

# 2. Run container
docker run -it --rm -p 9696:9696 fitness-prediction

# 3. In a new terminal, test the service on the link: http://localhost:9696/docs
The endpoint is predict

🎯 Usage

The deployed model can be used to predict fitness levels based on health and lifestyle parameters, making it suitable for:

• 🏥 Health assessment applications
• 💪 Wellness programs

---

Note: This dataset is synthetic and intended for educational purposes to practice data cleaning, feature engineering, and classification modeling.

❤️ Deployed Video ☁️ Fly.io