Yizhen Wang

Teaching Assistant for Data Management and Statistical Learning, supporting students in MySQL and NoSQL systems (Spark, MongoDB) and classical statistical methods, including Bayesian models.

● Designed and built an end-to-end ML solution for insurance claims in Python using 150k emails’ text data, boosting the production model’s weighted F1 score by 5.3% and increasing the model recommendation acceptance rate by 13% ● Built a text processing framework with Claude 3.5 on AWS Bedrock to clean and normalize email text (e.g., footer removal, lemmatization, stop-word filtering), improving data quality and boosting weighted F1 by 8% compared to baseline ● Proactively engineered a new feature with Latent Dirichlet Allocation (LDA) to capture thematic patterns, improving weighted F1 by 8.3% compared to baseline ● Evaluated 3 text embeddings (TF-IDF, GloVe, Word2Vec) across ML models (Logistic Regression, SVM, LightGBM, XGBoost) with Optuna hyperparameter tuning; further developed a BERT-based deep learning model, improving weighted F1 by 200% over baseline ● Performed model interpretability analysis with SHAP to quantify global and local feature contributions, identifying key drivers and increasing stakeholder confidence ● Presented model insights to non-technical stakeholders, bridging the gap between technical results and business applications, and led the design of a business-rule-aligned label selection logic, raising model recommendation acceptance rate by 13%

● Developed machine learning models to predict Returned Deposit Items (RDIs) from 1M+ consumer transaction records in Python, improving recall from 0.2 to 0.7 compared to the baseline model and enabling earlier fraud detection and faster check processing

● Analyzed 600K property sales records in Python and built ML models (XGBoost, Random Forest, Lasso), achieving R² > 0.85 and empowering clients to make data-driven pricing decisions in dynamic real estate markets

● Identified high-priority regions for traffic safety improvements through data analysis and delivered actionable insights through reports and weekly presentations, helping secure $20K in government funding and contributing to reduced traffic incidents

• Built an automated R Shiny dashboard for budget tracking, reducing manual work by 80% and improving data accessibility • Streamlined multi-source data workflows and collaborated with stakeholders to deliver real-time insights that accelerated decision-making