Machine Learning in Epidemiology with R — Part 2
R-Ladies Rome Tutorials
Registered Attendees (38 on Meetup + 28 on Luma)
On October 23, 2026, R-Ladies Rome hosted Machine Learning in Epidemiology with R — Part 2, a workshop led by Federica Gazzelloni.
Building on our September introduction to machine learning, this session moved from understanding how models learn to working through a clinical classification problem in R:
Can clinical measurements help us distinguish patients with and without heart disease?
Using data from 270 patients, the workshop connected saved random forest results with the steps needed to prepare, train, and evaluate a neural network. The task was to classify existing heart disease, rather than predict a future cardiovascular event.
🧩 Workshop Overview
The workshop drew on Wright et al.’s (2026) Machine Learning in Epidemiology and followed a practical workflow using R, Keras, and mlr3.
We started with the heart dataset and the predictions from a previously fitted random forest. Reading its correct classifications and mistakes gave us a reference for the next model, while keeping the focus on what predictions can tell us about the patients in the test data.
From there, the materials introduced an mlr3 classification task, reused the saved training and test groups, and prepared numerical inputs for a neural network. The Keras workflow covered building the network, fitting it, and examining training and validation loss before evaluating its predictions with mlr3.
Throughout the workflow, we returned to three questions:
- Which data are used to train the model, and which are kept for evaluation?
- What do changes in training and validation loss tell us?
- What can we conclude from comparing two models on the same test set?
The comparison provides a worked example, not a general ranking of model families. Scores from one dataset and one split need to be interpreted in that setting, and predictive performance does not establish causation.
🎥 Recording
🎬 Watch the Recording
Follow the recording alongside the teaching notebook to revisit the heart-data example, work through the code, and examine how model training connects to evaluation.
🧠 What You’ll Learn
By watching the recording and working through the materials, you will learn how to:
- frame heart-disease classification as a supervised learning task
- read saved random forest predictions and identify classification errors
- define a classification task and reuse training and test groups with
mlr3 - prepare numerical inputs for a neural network
- build and fit a neural network with Keras in R
- interpret training and validation loss
- evaluate predictions and compare results on the same test data
- recognise the limits of a single model comparison
📦 Resources & Materials
- Workshop repository: both parts, setup instructions, and code
- Part 2 session guide and installation instructions
- Part 2 teaching notebook
- Part 2 presentation (PDF)
- Data guide and source details
- Recorded validation checks and local results
- Part 1 recap and recording
The Part 2 guide also includes optional follow-up activities on trees and resampling, hyperparameter tuning, and synthetic data. These extend the main lesson for anyone who wants to explore further.
References
- Wright et al. (2026), Machine Learning in Epidemiology.
- Federica Gazzelloni (2025), Health Metrics and the Spread of Infectious Diseases: Machine Learning Applications and Spatial Modelling Analysis with R. CRC Press.
🔊 About the Speaker
Federica Gazzelloni is an actuary, statistician, data scientist, author, instructor, and organiser of R-Ladies Rome. Her work spans health metrics, statistical and actuarial modelling, machine learning, spatial analysis, and reproducible research. She is the author of Health Metrics and the Spread of Infectious Diseases: Machine Learning Applications and Spatial Modelling Analysis with R, published by CRC Press in 2025.
💜 About the Tutorial
This workshop is part of the R-Ladies Rome Tutorials series, where we focus on practical skills, reproducible workflows, and the reasoning behind the tools we use.
Across the two sessions, the aim was to make the learning process easier to follow: from parameters and loss functions to fitting models, reading predictions, and questioning what an evaluation supports. The notebooks and scripts provide a way to revisit each step at your own pace.
Thank you to everyone who joined and learned with us.
Keep learning and exploring, subscribe to our YouTube channel, and revisit past events on rladiesrome.org.