optuna-optimization

SaleFore AI: Ultra-accurate sales forecasting using ensemble ML (XGBoost, LightGBM, CatBoost) with RTX 4060 GPU optimization. Achieves 88-95% accuracy with advanced hyperparameter tuning.

This project was developed for the ML Engineering Postgraduate Program, where a classification machine learning model was built to predict whether a customer will subscribe to a term deposit after a marketing campaign.

A reinforcement learning trading agent that uses Proximal Policy Optimization (PPO) with automated hyperparameter tuning via Optuna to learn optimal trading strategies.

Understanding menstruation and cycle length using clustering, predictive modeling and model interpretability

A comprehensive framework for developing and backtesting quantitative trading strategies.

End-to-end ML project predicting NYC taxi fares using XGBoost + Optuna on a 33M row dataset | R² = 0.9851 | MAE = $0.66

A machine learning credit scoring

rsna_pneumonia_project

2024 한국인공지능융합기술학회 추계학술대회에 제출한 논문에 대한 연구 내용입니다.

Banking_ML_Project

This project explores Attention-Based Transformer Encoders to develop robust buy/sell classification models for financial time series. It addresses market non-stationarity and noise by combining De Prado-inspired preprocessing with a hybrid Transformer-LSTM architecture.

Credit risk prediction using machine learning, ensemble modeling, Optuna optimization, SHAP explainability, fairness analysis, and interactive deployment.

End-to-end fraud detection system on 590K IEEE-CIS transactions — XGBoost + Autoencoder ensemble (0.60% FPR), SHAP explainability, deployed via FastAPI + Streamlit.

Hourly Energy Consumption

This project implements a **Handwritten Digit Classification** system using the **MNIST dataset**. The model is trained to recognize digits from `0–9` based on grayscale images of handwritten characters. The project demonstrates the application of deep learning techniques for image recognition tasks.

Predicting telco customer churn with deep learning and advanced feature engineering on the Telco Customer Churn dataset.

This project implements a Fashion MNIST Classification system using the MNIST dataset. The model is trained to recognize Fashion objects like shirts,shoes,trousers etc. based on grayscale images of clothes. The project demonstrates the application of deep learning techniques for image recognition tasks.

CommentSense v1: Fine-tuned Twitter-RoBERTa transformer on 1M+ YouTube comments, deployed with FastAPI and Hugging Face, and integrated into a Chrome Extension for real-time sentiment analysis.

Production-grade credit risk ML pipeline — Optuna-tuned LightGBM (ROC-AUC 0.91), isotonic probability calibration, business-profit-matrix threshold optimization ($6.74M), SHAP global + local explainability, Fairlearn geographic fairness audit, and joblib-persisted inference system.

A Multimodal Regression Pipeline that predicts property market value using both tabular data and satellite imagery.