Data Science | Machine Learning

Technical Skills: Python, R, SQL, GCP

Support Skills: Git, Docker, Ubuntu, Spark

Education

• M.S., Statistics @ NCCU (Aug 2023)
• 2nd Specialty, Computer Science @ NCCU (Aug 2023)
• B.S., Mathematics @ NTUE (Jun 2021)

Work Experience

Data Scientist (Internship) @ The News Lens Research (Sep 2022 - Sep 2023)

• Developed automated analytic processes that reduce the time required from approximately 3 to 4 days to less than 30 minutes.
• Created user behavior data dashboards using Looker Studio and BigQuery, simplifying data visualization and enhancing accessibility.
• Led collaborations that significantly doubled company revenue within a year.
• Products:
  • SoundFlow: A Bert Based NLP Analytics Report on GCP
  • CDP Automated Analysis Program
  • Domohorn Wrinkle Customer Feedback Analysis
  • Australia Automobile Market Analysis
  • ShareParty User Behavior Analytics Reports

Project

Pic2Recipe: A Yolo based recipe recommender

Cloud Run | Streamlit
mAP50: 0.9| mAP50-95: 0.68
Using Python and Ultralytics to build a YOLO model capable of detecting food ingredients in a picture, and subsequently developing a recommender system to suggest users what kind of dishes they can cook.

WSDM KKBox's Music Recommendation Challenge

Top 25% solution on Kaggle
Using Python, XGBoost, Sklearn to build a music recommendation system.
ML techniques such as: User-item Matrix, PCA, Normalization, ...etc. were used in this project.

Solar Panel Dust Detection

Accuracy: 90%| F1-Score: 85%
Using Python, Selenium, Tensorflow(&Keras), and EfficientNet, I collect solar panel images through web crawling and perform data preprocessing tasks, including addressing imbalanced data, resizing, and augmentation. Subsequently, I employ various model architectures such as CNN, SVM, ResNet50, VGG19, and EfficientNetB3 to create a model for the detection of dirt or contaminants on solar panels.

Stroke Prediction By Deep Learning

Accuracy: 82%| Recall: 58%
Applying DNN by using Keras to predict whether a patient will get a stroke based on the input parameters like gender, age, various diseases, and smoking status.

Spotify Popularity Prediction

Using R to perform dimension reduction and machine learning methods like PCA, FA, Random Forest, XGBoost,... etc. to make a prediction of popularity of songs in Kaggle Spotify data.

Competition

1. IMBD 2022: Finalists (Nov 2022)
2. AI CUP Fall 2023: Ongoing (Oct 2023)
3. AI GO 2023: Ongoing (Oct 2023)

Certificate

• TOEIC Listening and Reading Test -- Score 835
• Datacamp Data Analyst with Python Track

Course

• Deep Learning: Fundamentals and Applications
• Pattern Recognition
• Data Science
• Statistical Computing and Simulation
• Advanced Mathematical Statistics
• Applied Regression Analysis
• Multivariate Analysis
• Sampling Methods

Links

Linkedln Medium