target-encoding

Lightning-fast data preprocessing and feature engineering for Python, built on Polars. 108 sklearn-style transformers for imputation, encoding, scaling, outlier clipping, discretization and feature selection, with Pipeline support and ONNX export for low-latency inference.

A python machine learning library for advanced feature extraction and interpretation.

A set of tools for machine learning (for the current day, there are active learning utilities and implementations of some stacking-based techniques).

ML system predicting Indian domestic airfares (R2 0.9165, 94.1% accuracy) and recommending cost-effective bookings, with validated ranking reliability

Fold-safe ensembling to the honest ceiling of Kaggle Playground S6E9, and a Darwin-Cage residual search with a power analysis. Nested OOF 0.946420; 47,679 evolved programs, 0 discoveries.

Production-ready ML pipeline demonstrating feature engineering, hyperparameter tuning, stacking ensembles and pseudo-label learning.

Welcome to the Diabetic Analysis Project repository! This project was created during my Data Science Internship at Exposys Data Labs. It showcases an end-to-end machine learning workflow for early diabetes prediction, from data loading and preprocessing to model building, cross-validation and model selection.

Kaggle Playground S6E9 (EV purchases, ROC AUC): nested target-encoding pipeline over synthetic-generator artifacts, 190+ measured experiments, ceiling analysis. OOF 0.9465.

Upstream feature_engine v1.9.4 plus one added OOFMeanEncoder (leak-free out-of-fold target encoding), proposed back to feature-engine/feature_engine as issue #1050. An upstream snapshot, not a standalone project.

Machine learning pipeline for predicting trucking load rates from lane, equipment, market, and temporal features, with training, prediction, and submission validation.

Kaggle Playground S6E10: airline passenger satisfaction (ROC AUC). Public LB top 4%, route features, in-fold target encoding, 278-model stack with nested CV.

Home Credit Risk Prediction (AUC 0.794 / Private LB Top 15%)

End-to-end regression framework featuring K-Fold median target encoding, Optuna hyperparameter optimization, cross-validation, and MAE-optimized ensemble blending using LightGBM and CatBoost.

Hybrid stacking system: [MLP, VAE, LTSM XGBoost, LightGBM, CatBoost, RandomForest] -> LogReg

Kaggle Playground S6E11: target encoding of exact values and digits on synthetic data. Data representation beat model tuning (ROC AUC 0.94732 -> 0.95149)

Encode Categorical Features based on Target/Class

🩺 Predict diabetes risk using an end-to-end machine learning pipeline, featuring advanced models and techniques for superior accuracy in the Kaggle competition.

Bank churn prediction (PS S4E1) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB ROC-AUC 0.89102.