Renal AML Growth Modeling

Research objective

I led an undergraduate team exploring whether machine learning could predict the growth of renal angiomyolipomas. We evaluated both regression for continuous growth rates and classification into high- and low-growth groups.

Pipeline and evaluation

I developed a transformation pipeline for 653 records and 706 initial features, including medically informed feature removal, missing-data handling, one-hot encoding, and leakage prevention. We compared neural networks, XGBoost, Gaussian processes, and support vector machines against clear naive baselines.

What we learned

The trained models did not consistently outperform simple baselines. Rather than overstate the results, we documented how limited, noisy data constrained the study. The project became an important lesson in rigorous benchmarking, intellectual honesty, and knowing when a dataset cannot support the desired conclusion.