🚀 AutoML Pipeline Project for Classification Task

This repository showcases an Automated Machine Learning (AutoML) Pipeline for a classification task (predicting "Survived"). It is designed to demonstrate end-to-end data science workflow automation, advanced ML techniques, and proficiency with key tools and libraries.

1. 🔄 Data Cleaning

   • Missing value imputation (median/mode)
   • Categorical encoding (One-Hot Encoding)
   • Feature scaling (StandardScaler)

2. ⚡ Feature Engineering

   • Custom feature transformations
   • Automated feature synthesis using Featuretools (Deep Feature Synthesis)

3. 📊 Rich Visualizations

   • Pre-training: Correlation heatmaps, target distribution, pairplots
   • Post-training: Confusion matrix, classification report, ROC curve
   • Visualizations are saved as .png files in the visualizations/ directory

4. 🎯 Hyperparameter Tuning

   • Optuna for hyperparameter optimization
   • Implements K-Fold Cross-Validation and prunes underperforming trials for efficiency

5. 🤖 Model Selection & Training

   • Multiple ML models (Logistic Regression, Random Forest, XGBoost, SVM, etc.)
   • Selects the best model based on validation performance

6. 📋 Experiment Tracking with MLflow

   • Logs model parameters, metrics (Accuracy, F1-Score, ROC-AUC), and artifacts
   • MLflow UI enables visualization and comparison of multiple experiment runs

1. Install dependencies:

   pip install -r requirements.txt

   Or individually:

   pip install numpy pandas scikit-learn featuretools optuna mlflow seaborn matplotlib xgboost

2. Place your raw data in data/raw/data.csv

   • Ensure the dataset contains a column Survived(or a final column that will be renamed to Survived). For your data just change name of your target with Survived.

3. Run the main script:

   python main.py

4. Explore the results:

   • 📁 Visualizations in the visualizations/ folder.
   • 📊 MLflow Tracking:

     mlflow ui --backend-store-uri sqlite:///mlflow.db

     Open http://localhost:5000 to view and compare experiment runs.

• Validation Accuracy: ~100%
• Test Accuracy: ~90%
• F1-Score and ROC-AUC metrics are also tracked in MLflow.

My data was pretty limited and seperable so high scores are expected.

• Every experiment run logs model parameters, metrics (Accuracy, F1-Score), and artifacts.
• All runs are stored in an SQLite MLflow database (mlflow.db).
• Start the MLflow UI:

  mlflow ui --backend-store-uri sqlite:///mlflow.db

  Explore experiments at http://localhost:5000.