Machine Learning in Epidemiology with R — Part II: From Trees to Neural Networks
Details
How do different machine-learning models learn from the same data, and how can we compare their predictions?
Join R-Ladies Rome for the second workshop in our Machine Learning in Epidemiology with R series, led by Federica Gazzelloni.
Part I introduced how models learn through loss and optimisation.
Part II, we’ll put these ideas into practice by training and comparing decision trees, random forests and a simple neural network using a heart-disease classification example.
We’ll follow a complete workflow in R: explore the data, define a prediction task, train models and evaluate their performance on observations they have not seen during training.
What we’ll cover
- Decision trees and random forests with `mlr3`.
- A simple neural network with Keras.
- Layers, activation functions, epochs and batch sizes.
- Training and validation loss.
- Model comparison using accuracy, sensitivity, specificity and ROC/AUC.
- Overfitting, cross-validation and the distinction between prediction and causation.
Who is this for?
This workshop is for people interested in applying machine learning to epidemiology and health data. Basic familiarity with R and RStudio is recommended. The session builds on Part I and begins with a short recap.
What to bring
A computer with R and RStudio installed. Setup instructions and workshop materials will be shared before the session.
Bring your questions and join us for a practical exploration of how different models learn—and how we assess their predictions.
