Artificial Intelligence in BME Final Project

Authors: David Chang and Mikey Komaiha

Date: April 2020

We evaluated different machine learning algorithms for predicting primary biliary cirrhosis patient outcomes.

This project used both unsupervised and supervised machine learning to predict patient survival time and status (alive or deceased) after a 10 year period.

Unsupervised learning used to explore data:

• Principal component analysis (PCA)
• K-means clustering

To predict patient survival time, we used the following models:

• Stepwise Regression [Performed best according to Pearson's correlation] --> Used subset of 12 noninvasive variables with log values substituted for albumin, bili, and protime
• Linear Regression
• Lasso
• Random Forest

To predict patient survival status, we used the following models:

• Logistic regression
• Random forest [Performed best according to Pearson's correlation]
• Support vector machine

Original data and reformatted data files contained in data

MATLAB live script (.mlx) and html output files contained in results. We standardized all the data and ran the code again for finalProjectZscore.html.

For full details, please view Final Report.pdf.

Original Research Paper: Prognosis in primary biliary cirrhosis: Model for decision making

Original Dataset: https://github.com/therneau/survival/blob/master/data/pbc.rda