Mini-AutoML Tool
Running a full ML pipeline (cleaning, model selection, evaluation) on a new dataset takes real setup time, even for simple use cases.
Technology Architecture & Stack
Built a lightweight Automated Machine Learning pipeline framework accessible via interactive Streamlit Web UI, CLI, or Python API. It automatically infers column datatypes, handles missing value imputation, scales numerical features, and one-hot encodes categorical variables. Evaluates multiple classification or regression baseline models (Logistic/Linear Regression, Random Forest, KNN, Decision Trees) using GridSearchCV for hyperparameter optimization, selects the top performer, and exports the winning model as a ready-to-deploy .joblib pipeline file.
Key Features & Functionality
- Automated feature engineering: dtype inference, imputation, scaling, and one-hot encoding.
- Automated model exploration across classification and regression tasks.
- Hyperparameter tuning using GridSearchCV cross-validation.
- Interactive Streamlit web app for real-time dataset ingestion, model evaluation, and inference on unseen test samples.
- Model artifact export: serializes winning pipelines into production-ready .joblib files.
System Architecture & Technical Design
Encapsulates scikit-learn Pipeline and ColumnTransformer objects into an end-to-end unified workflow that prevents data leakage during cross-validation.
Key Engineering Takeaways
Building abstract scikit-learn wrappers deepened insight into robust data pre-processing and hyperparameter space exploration.