Predicting Life Satisfaction Using Machine Learning and Explainable AI

Table of Contents

1. Introduction
2. Repository Structure
3. Highlights
4. Getting Started
   • Prerequisites
   • Installation
5. Dataset
6. Notebook Structure
7. Results and Insights
8. Explainable AI
9. Ablation Studies
10. Citation
11. Contact
12. License

---

Introduction

This repository accompanies the research article "Predicting Life Satisfaction Using Machine Learning and Explainable AI", published in Heliyon. The project demonstrates how advanced machine learning and explainable AI (XAI) techniques can predict life satisfaction with high accuracy. The dataset, sourced from the SHILD survey in Denmark, provides critical insights into factors affecting well-being. The study also explores the use of large language models (LLMs) for predicting life satisfaction, achieving significant results.

Publication Link: Heliyon Article

---

Repository Structure

├── Figures/                   # Contains visualizations used in the notebooks
├── LICENSE                    # License information
├── README.md                  # Repository documentation
├── Predicting_Life_Satisfaction.ipynb  # Main Jupyter Notebook

---

Highlights

• Achieved 93.8% accuracy and 73% macro F1-score for predicting life satisfaction.
• Used Recursive Feature Elimination with Cross-Validation (RFECV) to identify 27 key determinants of life satisfaction.
• Employed Explainable AI techniques to ensure interpretability and transparency of predictions.
• Explored Large Language Models (LLMs) like BERT, BioBERT, and ClinicalBERT to predict life satisfaction using natural language sentences.
• Conducted ablation studies on data resampling and feature selection techniques to optimize model performance.

---

Getting Started

Prerequisites

• Python 3.6+: Download here
• Jupyter Notebook: Install via pip:

  pip install notebook

Installation

Clone the repository and navigate to the directory:

git clone https://github.com/alifelham/Predicting-Life-Satisfaction-Using-Machine-Learning.git
cd Predicting-Life-Satisfaction-Using-Machine-Learning

Install required libraries:

pip install numpy pandas matplotlib scikit-learn seaborn missingno imbalanced-learn scikit-plot xgboost lightgbm

---

Dataset

The dataset is sourced from the SHILD (Survey of Health Impairment and Living Conditions in Denmark). It is publicly available under a CC0 1.0 Universal Public Domain Dedication license.

Dataset Link: SHILD Dataset

---

Notebook Structure

1. Data Importing and Preprocessing: Handles missing values, categorical encoding, and outlier management.
2. Exploratory Data Analysis: Visualizations and data summaries.
3. Model Building: Implements ML models such as Random Forest, XGBoost, and LightGBM.
4. Model Evaluation: Uses metrics like accuracy, F1-score, precision, recall, and AUC-ROC.
5. Results Visualization: Displays model performance and insights.
6. Explainable AI: Interprets predictions using XAI techniques.
7. Age Group Analysis: Examines primary determinants across different age brackets.

---

Results and Insights

Key Performance Metrics:

Model	Accuracy (%)	F1-Score (%)	Precision (%)	Recall (%)
Random Forest	93.8	70.6	72.0	69.3
Gradient Boosting	92.2	70.3	67.9	73.7
XGBoost	93.0	68.5	68.7	68.2
Ensemble (Best)	93.6	73.0	71.9	74.3

Insights:

• Health condition is the most critical determinant across all age groups.
• Dual data resampling (SMOTE + undersampling) improves both accuracy and F1-score.
• RFECV-based feature selection outperforms PCA-based approaches.

---

Explainable AI

Explainable AI was employed to ensure model transparency. The framework explains how each input feature contributes to the prediction, providing actionable insights for stakeholders like policymakers and healthcare professionals.

Example Visualization: Add visualizations here for model explanations or prediction thresholds.

---

Ablation Studies

Data Resampling:

The dual strategy of oversampling and undersampling led to significant improvements in model performance, achieving a balanced precision-recall tradeoff.

Feature Selection:

RFECV selected 27 key features, surpassing PCA in both accuracy and interpretability.

---

Citation

If you use this repository, please cite the following:

@article{alifelham2024lifesatisfaction,
    title={Predicting life satisfaction using machine learning and explainable AI},
    author={Alif Elham Khan, Mohammad Junayed Hasan, Humayra Anjum, Nabeel Mohammed, Sifat Momen},
    journal={Heliyon},
    year={2024},
    doi={10.1016/j.heliyon.2024.e31158}
}

---

Contact

For questions or collaboration, contact:

• Alif Elham Khan: alif.khan1@northsouth.edu
• Mohammad Junayed Hasan: mohammad.hasan5@northsouth.edu
• Humayra Anjum: humayra.anjum@northsouth.edu

---

License

This repository is licensed under the MIT License. See the LICENSE file for details.