Machine Learning Projects

• Home Credit Default Risk
  1. Developed k-fold LightGBM (Light Gradient Boosting Machine) model.
  2. Predicted repayment capabilities of each client.
  3. Achieved AUC (area under the curve) to 0.787 compared with Logistic Regression model 0.671 and Random Forest model 0.678.
  4. Training data has 307,511 observations and 122 features and testing data has 48,744 observations and 121 features.
• NLP Sentiment Analysis with Emoji Labels
  1. Converted training/testing features into vector list using pre-trained GloVe model.
  2. Inserted a keras Embedding layer given word to vector mapping.
  3. Build and trained a 2 layer LSTM network.
  4. Achieved sentiment classification for sentences and labeled with emoji.
• Game Recommedation System
  1. Scrapped user data (17.3MB) and app data (510.2MB) from Steam.
  2. Sotred app data in MySQL database.
  3. Trained content-based model and item-based model.
  4. Created a website providing 5 recommendations for each user in list.