Projects
Selected coursework and data analysis projects.
Survival Analysis on Telco Customer Churn Data
In this project, I followed the official survival analysis tutorial and implemented a full workflow including data preprocessing, Kaplan–Meier survival estimation, Cox proportional hazards modeling, and Accelerated Failure Time model comparison.
The analysis focused on customer churn as a time-to-event problem and showed that service-related features such as online backup and tech support were strongly associated with lower churn risk.
Text-to-SQL Error Analysis with LLMs
I constructed a MySQL database from the Telco Customer Churn dataset and tested whether an LLM could generate correct SQL from natural language questions. The experiment included both successful cases and failure cases.
The results showed that LLMs can handle some simple aggregation queries, but may fail due to schema misunderstanding, malformed structured output, or incorrect subgroup aggregation logic.