ChenglongChen/kaggle-HomeDepot

Employs a multi-stage ensemble combining XGBoost, Keras neural networks, and RGF regressors with extensive feature engineering—including Word2Vec/GloVe embeddings, text similarity metrics, and domain-specific dictionaries for spelling correction and color data. Uses hyperopt for automated hyperparameter tuning across learners, stacking predictions across cross-validation folds for 2nd/3rd level ensemble aggregation. Feature selection combines regex-based manual curation with correlation filtering, leveraging thousands of candidate features derived from product titles, search queries, and descriptions through NLP preprocessing.

466 stars. No commits in the last 6 months.