youssefsultan / predict-bodyfat-using-catboost-webapp

This repository is comprised of the Exploratory Data Analysis of the Body Fat data set from Kaggle. The feature engineering, hyperparameter tuning, and model training of the model. With comparative outlooks on the prediction vs actual results to understand and determine model accuracy.

Body Fat Percentage Predictor: gathers body attributes to make a prediction on your bodyfat percentage based on previous data of recorded participants utilizing hydrostatic weighing. Additionally, the model has been deployed end to end using Streamlit, for testing and further research for applicability.

• Original Dataset: Kaggle Link
  
• EDA: BodyFat_Prediction.ipynb
  
• Deployment: Streamlit Webapp
  
• Tableau Dashboard: Tableau Public or Download
  

Application

• Variable correlation testing
• Variable distribution analysis
• Outlier detection and feature engineering
• Model hyperparameter tuning and training

Results

• Using various learning rates, tree depths and number of trees this model's accuracy shows a positive linear relationship
• Achieves a Mean Absolute Error of 134.66 with normal distribution comparing the prediction and actual values
• CatBoost was used simply due to speed and optimization in comparison to XGBoost or LightGBM