https://github.com/ata-turhan/titanic-survival-prediction
A comprehensive solution for the Kaggle Titanic Challenge, featuring advanced data exploration, feature engineering, model training, and explainable AI techniques. Includes Logistic Regression, RandomForest, XGBoost, and Stacked Ensembles with SHAP and permutation importance for model interpretability.
https://github.com/ata-turhan/titanic-survival-prediction
classification kaggle python shap xgboost
Last synced: 7 months ago
JSON representation
A comprehensive solution for the Kaggle Titanic Challenge, featuring advanced data exploration, feature engineering, model training, and explainable AI techniques. Includes Logistic Regression, RandomForest, XGBoost, and Stacked Ensembles with SHAP and permutation importance for model interpretability.
- Host: GitHub
- URL: https://github.com/ata-turhan/titanic-survival-prediction
- Owner: ata-turhan
- License: mit
- Created: 2024-10-05T17:49:46.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2024-11-20T03:47:15.000Z (12 months ago)
- Last Synced: 2025-04-09T23:47:57.972Z (7 months ago)
- Topics: classification, kaggle, python, shap, xgboost
- Language: Jupyter Notebook
- Homepage:
- Size: 1.75 MB
- Stars: 11
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# **Titanic Survival Prediction**
This repository contains a comprehensive solution for the Kaggle **Titanic - Machine Learning from Disaster** competition. Using machine learning techniques, we predict the survival of passengers based on key features like age, gender, class, and more. This project also includes **Explainable AI (XAI)** techniques to understand model predictions.
---
## **Project Overview**
- **Goal**: Predict whether a Titanic passenger survived or not, using features such as age, gender, ticket class, and more.
- **Techniques**:
- Data Cleaning and Feature Engineering.
- Model Training and Hyperparameter Tuning.
- Model Stacking for improved performance.
- Explainable AI using SHAP, permutation importance, and feature importances.
---
## **Features**
- **Extensive Data Exploration**: Visualizations and insights from the dataset.
- **Feature Engineering**: New features such as `AgeGroup`, `FareBin`, `FamilySize`, `IsAlone`, and `Title`.
- **Model Training**: Logistic Regression, RandomForest, XGBoost, MultiLayer Perceptron, and Stacked Ensembles.
- **Explainable AI**: Insights into model predictions with SHAP values, permutation importance, and feature coefficients.
- **Kaggle Submission**: Prepares predictions in the required format for submission.
---
## **Setup Instructions**
### **1. Clone the Repository**
```bash
git clone https://github.com/ata-turhan/titanic-survival-prediction.git
cd titanic-survival-prediction
```
### **2. Install Dependencies**
Ensure you have Python 3.8+ installed. Install the required packages using:
```bash
pip install -r requirements.txt
```
### **3. Run the Notebook**
Launch Jupyter Notebook and open `notebooks/titanic_predicting_survivors.ipynb`:
```bash
jupyter notebook
```
---
## **Key Results**
| **Model** | **Accuracy** | **F1-Score** |
|-----------------------|--------------|--------------|
| Logistic Regression | 80.45% | 0.80 |
| RandomForest | 82.68% | 0.82 |
| MultiLayer Perceptron | 79.33% | 0.79 |
| XGBoost | TBD | TBD |
| **Stacked Model** | **83.50%** | **0.83** |
---
## **Explainable AI (XAI)**
This project uses several techniques to explain model predictions:
1. **Logistic Regression Coefficients**:
- Visualizes the top features affecting survival.
2. **RandomForest Feature Importances**:
- Highlights features most impactful for survival predictions.
3. **Permutation Importance**:
- Explains predictions of the MLP model by permuting feature values.
4. **SHAP Values**:
- Detailed explanations for RandomForest predictions with SHAP plots.
---
## **Kaggle Submission**
The repository includes the final predictions in the `outputs/` directory:
- **File**: `submission.csv`
- **Format**:
```csv
PassengerId,Survived
892,0
893,1
...
```
---
## **Future Improvements**
- Explore advanced stacking techniques
- Incorporate additional external datasets for richer feature engineering.
---
## **License**
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.