Distance Walked Prediction

Overview

This project predicts daily distance walked in kilometers from health and lifestyle features such as step count, sleep duration, blood pressure, occupation, and BMI category.

Problem Statement

The goal is to estimate walking distance from health-related indicators and compare a small set of regression models on the same prepared feature set.

Dataset

Source: Kaggle-derived health activity dataset provided in this repository. The original source was not documented in the project files.

Technologies Used

• Python
• pandas
• numpy
• seaborn
• matplotlib
• scikit-learn

Project Structure

• data/raw/ stores the source dataset used by the notebook and script.
• notebooks/ contains the exploratory and modeling notebook.
• src/ contains the reusable Python workflow.
• images/ contains project figures used in the README.
• README.md documents the project.
• requirements.txt lists the Python dependencies.
• .gitignore excludes local Python and notebook artifacts.

Workflow

1. Load the health dataset from data/raw/Health_dataset.csv.
2. Clean missing values and derive Distance_walked(km) from daily steps.
3. Parse blood pressure into numeric features and encode categorical columns.
4. Split the data into train and test sets and scale features with RobustScaler.
5. Train Linear Regression, polynomial SVR, and Random Forest models.
6. Compare MAE, RMSE, and cross-validated R2, then visualize the results.

Results

The original analysis showed Random Forest as the strongest model overall, with SVR close behind and Linear Regression trailing on error and fit quality.

Future Improvements

• Test additional regression models while keeping the same feature pipeline.
• Add reproducible experiment tracking and saved model artifacts.
• Separate notebook exploration from a command-line training entry point.

Visualization