Daniel Wang
Pronouns: He/him
Research Mentor(s): Maria Han Veiga
Research Mentor School/College/Department: Mathematics / LSA
Program: CG
Session: Session 5 (2:40pm – 3:30pm)
Authors: Maria Veiga, Daniel Wang
Abstract
The Brazilian Peritoneal Dialysis Multicenter Study (BRAZPD) is designed to continuously monitor the peritoneal dialysis (PD) reality in Brazil. This dataset collects the data associated with the renal failure on the sample and the comorbidities caused by the renal failure for 75 months. The goal of this project is to use the baseline classification model to predict the patient outcome and using SHAP values to explore the interpretability of the model based on the BRAZPD dataset. One drawback of the BRAZPD dataset is that it has some artifacts so the first stage of this project has been data cleaning and normalization. In the dataset, parts of the sample didn’t last through the duration of 75 months. Checking with the duration of the study for each individual in the sample, ending points have been determined for data cleaning. Any data that contradicts with its corresponding following up endpoint has been set to none. After the initial cleaning, normalization has also been applied to the dataset for a later use in visualizing the data. Visualization of the dataset not only shows the distribution of each feature associated with the PD study, but also the trend that the number of participants decreases over time as a result of, mainly, their death. The next steps will be to set up a baseline classification model on patient mortality (binary classification) and patient cause of death (multiclass classification) and compute SHAP values to gain a better understanding of the data-driven model.
