Table of Contents

1. Explainable Digital Currency Candlestick Pattern AI Learner

IEEE Xplore | Conference Video

• The paper gets accepted at the Knowledge and Smart Technology (KST) 2022 International Conference, hosted by IEEE.
• Given the black-box of deep neural network (DNN), the study aims to expose and enhance the explainability of a DNN-based classic trading strategy, candlestick pattern recognition, and makes an acceptable justification for traders in the cryptocurrency market. For more details, please refer to the GitHub, conference video and full paper.

Skills: Explainable Artificial Intelligence, Data Augmentation, Adversarial Attacks, Convolutional Neural Network, Candlestick Patterns, Time-Series Encoding, Financial Vision

2. Machine Learning Notes

• It is a lecture note made for the machine-related courses offered in the graduate schools of Soochow University. It covers multiple machine learning methods, including EDA, Bayesian, Clustering, Decision Tree, Dimension Reduction, Ensemble, KNN, Logistic Regression, Neural Network, Regression, SVM, Validation. Each method has note of concept and demo of source code.
• The Final project aims to predict S&P 500 index’s Close price ten days ahead, using daily features (Open, High, Low, Close, Volume) of each day, including the current day and the past 30 days. Noticeably, it applies cross-validation to tune hyperparameters. To avoid overfitting and maintain the model’s robustness, it chooses hyperparameters which has max value of mean test score + 1.96 * std test score in cross-validation.

3. Deep Learning-Based Candlestick Pattern Recognition System

To take advantage of the candlestick pattern recognition to a great extent and support the trader, we build an AI-based system that can absorb professional traders' domain knowledge and excel in visual recognition. Moreover, the system can automatically execute the procedure and have the customized user interface. And the user of traders can avoid the human error and make investment more efficiently.

Skills: Deep Learning, Computer Vision, Python, Django, Javascript, Bootstrap, Financial Domain

4. Deep Learning-Based Recommendation APP for Taiwan Tourist Attractions

Project Report

• This project won the championship in the National College Open Data Artificial Intelligence Competition.
• The project aims to build an iOS app that provides recommendations for tourist attractions in Taiwan to fulfill users' travel needs while visiting Taiwan. When users capture any interesting scenes around the world, and input them into the app, the system will utilize a deep learning model to recognize and recommend Taiwan tourist spots that are most like the captured scene.

Skills: Deep learning, Transfer learning, iOS Core ML 2.0, Image data annotation, Google Cloud computing, Data augmentation

5. Cathlife Critical Illness Insurance Purchasing Data Analysis

Analysis Document

1. The project won the 34th place out of 300 nationwide in the Cathay Life Big Data Competition when I represented the lab to participate during internship at IF.Lab.
2. The project aims to utilize customers' historical data to predict whether existing customers will purchase critical illness insurance policies within a specific timeframe. The model will help identify customers with higher insurance needs and provide sales professionals with a method to target potential customers.
3. The analysis focuses on the domain-based data analysis, and also places lots of effort on the Exploratory Data Analysis (EDA) and data cleaning. And, those pre-processing methods and model construction also covers multiple machine learning methods.

Skills: Machine Learning, Python, Exploratory Data Analysis (EDA), Data Cleaning, Domain-based Data Analysis

6. Ensemble Machine Learning on Crowdfunding Data for Project Status and Amount Pledged Prediction

1. The project aims to evaluate the success status and the final pledge amount of a launched project at the early stage in Kickstarter and Indiegogo, which are two world-famous crowdfunding platforms.
2. To predict the success status and the final pledge amount of a launched project, I apply the Stacking, an ensemble machine learning method, to make prediction. Furthermore, I also train multiple machine learning models with hyperparameter tuning and compare the performance of the Stacking model with those models. For more details of the project, please refer to the GitHub.

Skills: Machine Learning, Python, Exploratory Data Analysis (EDA)

7. Kaggle's House Prices Prediction

The goal of the project is to predict the final sale prices of houses. The dataset contains various features of residential home, such as the number of bedrooms, the size of the lot, the neighborhood, and many others. The project covers much feature engineering, involving handling missing values, transforming variables, creating interaction terms. Additionally, it applies Stacking, ensemble learning, and hype-tunes multiple machine learning model to predict and make comparison.

Skills: Machine Learning, Python, Feature Engineering, Data Cleaning, Domain-based Data Analysis

8. Teaching Notes for Financial Innovation Course at National Taiwan University

Teaching Note

1. It is a teaching note I made for a deep-learning-related class I taught in the financial innovation course, offered by Prof. Yun-Cheng Tsai in National Taiwan University, when I was a lecturer. The class is about applying deep-learning-based visual recognition to financial trading strategy, like candlestick pattern recognition.
2. The class covers the concept of trading strategy and the deep learning model and how to put the idea into practice. The practice includes data processing, time series data encoding and Convolution Neural Network modeling and predicting. For more details, please refer to the teaching note and source code.

Skills: Deep Learning, Data Processing, Python, Financial Investment