Misinformation Classification & Model Evaluation
Consolidated and analyzed 21,724 news articles; built reproducible preprocessing and modeling workflows and evaluated logistic-regression and hybrid models, achieving validation AUC up to 0.862. Diagnosed potential data leakage in a transformer-based modeling workflow, demonstrating model-risk awareness and emphasis on defensible validation rather than headline performance. Extended the research to LIAR2 political-claim data, emphasizing data quality, feature construction, classification methodology, and transparent comparison of modeling choices.