AutoML Framework – Meta-Dataset Driven Machine Learning Automation

This project presents a Meta-Learning powered AutoML framework that automates the entire ML pipeline — from dataset ingestion and preprocessing to intelligent model selection, hyperparameter optimization, and deployment.

Unlike conventional AutoML systems that rely heavily on trial-and-error searches, this framework introduces a Meta-Dataset Repository of dataset characteristics and past model performances to guide optimal model recommendation.

🚀 Key Features

• Meta-Dataset Repository
  • Curated from OpenML with 187 classification & 94 regression datasets.
  • Captures dataset size, feature types, distributions, class imbalance, and historical model performance
• Automated Data Preprocessing
  • 🧩 Missing Data Handling: Detects MCAR, MAR, MNAR patterns; applies Mean/Median, KNN, or Modified MICE imputation accordingly

  • 

  • 🔤 Semantic Normalization: Resolves spelling variations, abbreviations, and synonyms using NLP & Fuzzy Matching (FuzzyWuzzy + WordNet).
  • 📏 Adaptive Scaling & Normalization: Chooses between Min-Max, Standard, or Robust scaling based on dataset skewness, kurtosis, and outliers.
  • 🔄 Feature Encoding: Efficient one-hot encoding with cardinality management.
• Target Column & Task Identification
  • Identifies the target column using Regex, NLP embeddings, synonym matching, and fuzzy similarity.
  • Dynamically determines whether the task is classification or regression with >96% accuracy
• Meta-Learning Driven Model Selection
  • Uses KNN-based dataset similarity for finding optimal models from the repository.
  • Recommends both model type & tuned hyperparameters based on historical results.
• End-to-End Pipeline Automation
  • Automatically trains, validates, and deploys the best-suited model.
  • Packages trained models into ready-to-use inference scripts.

🛠️ Tech Stack

• Frontend: Streamlit (for dataset upload & interactive exploration)
• Backend: Python (Pandas, NumPy, Scikit-Learn)
• Meta-Learning: KNN-based similarity for dataset-to-model mapping
• NLP & Semantic Processing: SpaCy, WordNet, FuzzyWuzzy

📸 Demonstrations

⚙️ Installation & Setup

Clone the Repository

https://github.com/Ris15hab/AutoML.git
cd AutoML-MetaDataset

Install Dependencies

pip install -r requirements.txt

Run the Streamlit App

streamlit run app.py

Upload your dataset (CSV format) and follow the interactive workflow.

📊 Performance

• Meta-Dataset Scale: 187 classification + 94 regression datasets.
• Imputation Performance:
  • MCAR → Mean/Median (94.2% detection accuracy)
  • MAR → KNN (87.6%)
  • MNAR → Modified MICE (79.3%)
• Target Column Detection: 96.5% accuracy across diverse datasets.