A curated collection of machine learning projects including real-world forecasting, structured competitions, and classical problems.
-
Description:
A Kaggle-style competition hosted on DACON, predicting building-level electricity consumption. -
Focus:
Feature engineering (time-based, building type), ensemble models, Optuna tuning. -
Highlights:
- Explored clustering-based splits and weighted VotingRegressor.
- Achieved best SMAPE on public leaderboard around 6.99 ~ 7.00.
-
Description:
Submissions and model experiments for the SCU AI Competition. -
Focus:
Feature engineering, clustering, ensemble modeling. -
Result:
🏆 Best AUC: 0.897686 (VotingClassifier with engineered features)
-
Description:
A separate track of the SCU AI Competition held in 2025. -
Focus:
Structured daily experiment logs with feature engineering and model iteration.
Includes Kaggle-style reproducibility and checkpointed submissions.
-
Description:
Classic binary classification task on Titanic dataset. -
Focus:
Baseline experiments, ensemble methods, reproducible ML pipeline practice.
- Python, Jupyter Notebook
- scikit-learn, XGBoost, LightGBM
- pandas, numpy, matplotlib, seaborn
- Optuna (hyperparameter tuning)
This repository is a hands-on playground for practicing and validating machine learning workflows.
Each project includes steps like EDA, feature engineering, model training, evaluation, and iteration.
It serves as a learning record, with code structured for reusability and experimentation.
- Projects are standalone and modular.
- Repo is continuously updated as new experiments are added.
- Some models (e.g., SCU submissions) include day-by-day experiment tracking.