npm

npm Health Model

The idea behind this model is to predict based on data obtained from GrimoireLab if a project is adding risks to organizations that are consuming it.

This is what you will find in this directory:

• data/: The data used to train and test the model is available under. The data was collected on 2026-06-19
• notebooks/: Jupyter notebooks to compute the model based on the data
• docs/: A description of the process to allow future improvements

A brief description of the process

These are the main steps we followed to build the model:

• 1. We selected a small set of projects. An expert evaluated them as either healthy or unhealthy.
• 1. (1') In parallel, we calculated a bunch of metrics with GrimoireLab. Not all the metrics were relevant and some of them were redundant, so we selected the ones that added the most value.
• 2. We put everything together—the metrics and the evaluation of the expert—and trained the model. The result of the training was an equation with weights for each of the metrics.
• 3. The accuracy of the model was tested. If it looked bad, it was time to iterate; if it was good enough, we were done.