Automated Feature Engineering & Model Selection for Tabular Data

Description

This project implements a basic AutoML pipeline in Python designed to automate parts of the machine learning workflow for tabular datasets. It focuses on:

• Automated feature selection/dimensionality reduction using techniques like RFE and PCA.
• Automated training and evaluation of different machine learning models (Decision Trees, SVM, Neural Networks).
• Comparing the performance of different preprocessing/model combinations across various datasets.

The goal is to streamline the process of finding effective preprocessing strategies and models for structured data.

Features

• Loads CSV datasets.
• Performs basic preprocessing (imputation, scaling, encoding).
• Applies optional feature selection (RFE) or dimensionality reduction (PCA).
• Trains and evaluates Decision Tree, SVM, and a simple Neural Network.
• Compares results across different preprocessing steps and models.
• Outputs results to a CSV file.

Technologies Used

• Python 3.x
• Pandas
• NumPy
• Scikit-learn
• TensorFlow (Keras)

Setup

1. Clone the repository:

   git clone <your-repository-url>
   cd <your-repository-name>

2. Create a Python environment (recommended):

   • Using Conda:

     conda create -n automl_env python=3.9 # Or another version
     conda activate automl_env
     conda install pandas numpy scikit-learn tensorflow matplotlib seaborn -c conda-forge

   • Using pip and venv:

     python -m venv venv
     source venv/bin/activate # On Windows use `venv\Scripts\activate`
     pip install pandas numpy scikit-learn tensorflow matplotlib seaborn

Usage

1. Configure Dataset Paths:

   • Open the main script (automl_pipeline.py or your chosen filename).
   • Locate the CONFIG dictionary near the top.
   • IMPORTANT: Update the "path" values within CONFIG["datasets"] to point to the actual locations of your dataset CSV files (e.g., Titanic, Credit Card Fraud).

2. Run the Pipeline:

   python automl_pipeline.py

3. Check Results:

   • The script will print progress to the console.
   • A summary of the results will be saved to automl_pipeline_results.csv (or the filename specified in CONFIG["results_file"]).

Team Members

• Khushi Choudhary
• Snehitha Gorantla
• Zakir Elaskar