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.

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

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

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

rsna_pneumonia_project

Banking_ML_Project

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.

2024 한국인공지능융합기술학회 추계학술대회에 제출한 논문에 대한 연구 내용입니다.

Hourly Energy Consumption

This repository contains a comprehensive deep learning solution for Alzheimer's Disease Classification using state-of-the-art DenseNet architectures optimized with Optuna hyperparameter tuning. The project implements multiple DenseNet variants for classification of Alzheimer's disease stages from brain MRI images.

Leveraging XGBoost to predict whether a customer will subscribe to a bank's term deposit

A machine learning credit scoring

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.

Predicting student exam scores using LightGBM and CatBoost with advanced feature engineering | Kaggle Playground Series S6E1 | RMSE: 8.73

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

This study proposes a deep learning-based object detection framework utilizing YOLOv11 to automate the identification and classification of three common dental lesion types which are caries, gingivitis, and white spot lesions, using high-resolution intraoral photographic images.

Création d'un pipeline pour prévoir la consommation électrique d'une presse à balle

Implementation of GNN - Node embeddings, classification & Link Prediction on CORA & soc-hamsterster dataset

Kaggle Playground Series - Season 5, Episode 5

Loan default prediction notebook using traditional machine learning models and LightGBM. Tackling imbalanced financial data and evaluating performance with ROC-AUC.

A curated collection of machine learning and deep learning notebooks — classification, regression, CV, autoencoders, NLP, and time series forecasting with TensorFlow, PyTorch, and Ray Tune.

The final structure of my thesis project (notebooks and files still needs some polishing).

Customer churn prediction using XGBoost, LightGBM, and CatBoost with ensemble methods and Optuna hyperparameter tuning

Kaggle Playground Series - Season 5, Episode 7

AI-powered anemia detection with classical ML, refined datasets, and explainable predictions using SHAP.

Advanced credit card fraud detection using CatBoost + SHAP achieving 81% PR-AUC on highly imbalanced data (1:577 ratio), with production-ready pipeline, explainable AI, and quantified business impact ($2.1M annual savings)

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.

Generic template for building predictive models and optimization workflows that address the dual objectives of winning deals and ensuring profitability.

The approach extracts structured features, generates hybrid MiniLM + TF-IDF embeddings, and builds numeric/categorical features. An ensemble of 6 transformer regressors fuses multimodal inputs to predict log-prices, using hybrid embeddings and residual fusion for higher accuracy.

A modular AutoML framework for text classification using the IMDB dataset. The project compares CNN and RNN architectures for sentiment analysis and leverages Optuna for hyperparameter optimization. Built with TensorFlow/Keras, the pipeline is designed to be reusable, and extensible.