AnalysisLin CRAN R Package

  • Developed a personal exploratory data analysis (EDA) R package, AnalysisLin. It has been uploaded to the CRAN repository.
  • A quick and effective data exploration toolkit. It provides essential features, including a descriptive statistics table for a quick overview of your dataset, interactive distribution plots to visualize variable patterns, Principal Component Analysis for dimensionality reduction and feature analysis, missing value imputation methods, and correlation analysis

CRAN Website

Package Vignette

Residual Neural Network (ResNet) Project

  • Built a ResNet model, achieving 94% on the CIFAR-10 dataset for image classification
  • Integrated a self-attention layer into ResNet, utilizing it to forecast stock price movement (loss, gain, or flat) in the next 5 days with an accuracy of 42%, which is higher than the stochastic indicator’s accuracy of 33%.

ResNet Report.pdf

Code

CSE 538 NLP Term Project

  • Built ChatGPT2 and T5 LLM.

CSE 538 Project_Report.pdf

Code

Principal Component Analysis (PCA) vs Auto-encoder (AE) for Handwritten Digit Images from USPS

  • Conducted PCA by singular vector decomposition (SVD) and Auto-encoder(AE) with weights sharing, without weights sharing, and with more layers and nonlinear functions, to handwritten digit image from USPS
  • Compared the Frobenius norm error between the input matrices and reconstructed ones. Gained an understanding of the effect of different methods on feature dimensionality reduction.

Code

Longstaff-Schwartz method for pricing American option

  • Implemented the Longstaff-Schwartz method for pricing American options using Python.

Code

Insurance Cost Regression

  • Applied the Decision Tree-based methods - Random Forest and XGBoost to demonstrate that the variable smoker is the most significant factor in determining insurance charges.

Code

The Impact of COVID-19 To the Profit of Fortune 100 Companies

  • Performed a paired t-test to show there was not a significant difference in profit of Fortune 100 companies before the pandemic(2019) and after the pandemic (2021)
  • Performed a multiple linear regression to test which variables, such as revenue, asset size, market capitalization, size of the workforce, and the headquarters state political party, were responsible for the change in profit of those companies during these periods. Variable selection was performed based on the adjusted R squared.

Report.pdf

As well as many other unlisted projects for enhancing data analytical and coding skills.