My credit-card default prediction project compared KNN, Random Forest, and neural-network classifiers on historical account data. A tuned Random Forest reached 0.7771 ROC-AUC on a held-out test set, and I documented the trade-offs between identifying defaults and incorrectly flagging creditworthy customers.
For a yeast gene-expression project, I prepared and analyzed a high-dimensional dataset, then benchmarked several classifiers using leakage-conscious cross-validation. Regularized Logistic Regression on PCA-reduced features achieved approximately 51.5% cross-validated accuracy, outperforming the VAE-based approach; I also examined predictive-gene candidates and documented the study’s limitations.
I implemented and evaluated two approaches to facial verification: a custom CNN and a frozen FaceNet embedding pipeline. Neither met the specified accuracy and false-acceptance requirements, so my technical report distinguished prototype feasibility from deployable security and outlined areas for further testing.
In my SQL data-quality project, I used Python and PostgreSQL to investigate exception records, repair eligible service-request data, and verify updates. I’m currently a Retail Associate at The Home Depot, supporting customers and team members while adapting to changing operational needs.
