Obesity-Levels-Prediction

This project focuses on predicting obesity levels based on lifestyle, eating habits, and physical condition using machine learning.

Overview

The goal of this project is to analyze factors influencing obesity and build a predictive model for multi-class classification.

The workflow includes:

• Data preprocessing and feature engineering
• Exploratory Data Analysis (EDA)
• Model training and evaluation

Project Highlights

• Compared multiple machine learning models for multi-class classification
• Implemented both classical ML models and neural networks
• Applied feature engineering and encoding techniques
• Identified the best performing model (XGBoost)

Dataset

The dataset includes demographic data, eating habits, and lifestyle features used to predict obesity levels.

Models Used

• Logistic Regression
• Decision Trees
• Random Forest
• Support Vector Machines (SVM)
• XGBoost
• Neural Networks (RSNNS, nnet)

Results

Best Model

XGBoost achieved the best overall performance:

• Accuracy (test): 96.9%
• Cross-validation accuracy: 97.0%
• AUC: ~0.998

Model Comparison

• Random Forest: 95.95% accuracy, AUC: 0.999
• Decision Tree: 92.38% accuracy
• SVM: 81.67% accuracy (struggled with complex class boundaries)

Neural Networks

• nnet: 95% accuracy, Kappa: 0.9416
• RSNNS: 94.29% accuracy, Kappa: 0.9333

Key Insight

Models struggled most with adjacent classes (Overweight Level I vs II), while obesity classes were classified with high accuracy.

Installation & Usage

1. Clone the repository

git clone https://github.com/alaszmigiel/Obesity-Levels-Prediction.git
cd Obesity-Levels-Prediction

2. Open Jupyter Notebook

jupyter notebook

3. Open the file

Obesity_Levels_Prediction.ipynb

4. Run all cells

The notebook includes all required steps and installations if needed.

Technologies Used

• Language: R
• Data Processing: dplyr, tidyr
• Visualization: ggplot2, corrplot
• ML: caret, xgboost, kernlab
• Neural Networks: RSNNS, nnet
• Evaluation: pROC, MLmetrics