Classifying Fetal Health Conditions

Project Goal

The goal of this project was to create a multinomial logistic regression model to predict whether there was cause for concern regarding the fetal health of a child in the mother’s womb.

Cardiotocography is a technique used to monitor the fetal heartbeat, fetal movements, and the uterine contractions during pregnancy and labor. The machine used to perform the monitoring is called a cardiotocograph. The data set used in this study consists of a plethora of measurements produced by Cardiotocograms (CTGs), the machines used to perform Cardiotocography. The aim was to use these measurements as predictors in our model to accurately classify the health status of a child in the womb.

The Notebook

The the R Markdown document that details the process I took to complete this task can be viewed by clicking on the notebook link at the top of this page. The code can be found by clicking on the code link above as well.

What I Learned

I had many goals I wanted to achieve when I started this project, and they are outlined as such:

• Practice fitting the multinomial logistic regression model
• Practice balancing data sets via over/under-sampling class stratified observations
• Applying Principal Component Analysis (PCA) to a set of predictors/features
• Use visualizations and computed metrics to assess model fit
• Practice using the box package for explicit package and function dependencies (instead of using library() calls)

Some of these goals, such as using the incredible box package and creating various visualizations are code related. I enjoy learning about and using new packages, or packages that are at least new to me. I have used box in Shiny apps before, but it is relatively new to me when using it in a statistical modelling notebook setting. I believe it adds just as much value to these R Markdown notebooks as it does to Shiny apps, and using it to enforce clarity and reduce function namespace clashes is invaluable. I will continue to use it in every project I work on.

The other goals are mainly oriented towards using and interpreting more complex statistical concepts. Logistic regression is a tried-and-true distribution based statistical modelling technique and its multinomial extension is something that opens up new possibilities. Often, when a categorical response variable has more than two classes, modelers will merge classes to ensure binary logistic regression is usable. This obscures some of the true relationships within the data, however. The multinomial extension preserves the original relationships within the data and offers greater insight. It is an invaluable statistical tool that I can now add to my arsenal.

Applying class balancing to a multinomial response variable, as well as using Principal Components Analysis (PCA) in this setting was good practice for me as a data modeler. There are always trade-offs when using these techniques, and applying them to the fetal health data set has reinforced my understanding of those trade offs.