Playlist "Swiss Python Summit 2026"
Is Your Data Lying? Bayesian Data Credibility
Datasets are often treated as ground truth, even though they rarely are. Labels can be noisy, measurements can be imprecise, user segments can be under-recorded, and dashboards can hide uncertainty behind clean-looking numbers. Teams then often build models, metrics, and decisions on top of data whose reliability we have never actually measured.
This talk shows how Bayesian modelling can make data credibility explicit. Using simple, realistic examples, we will model uncertainty in labels, measurements, and sampling bias; distinguish signal from noise; and identify parts of a dataset that should not be trusted at face value. The goal is to employ Bayesian statistics as a practical tool for asking: where is this dataset solid, where is it fragile, and where might it be misleading us?
Attendees will leave with a practical framework for measuring data trustworthiness, and concrete examples of how Bayesian methods can make uncertainty visible before it starts to contaminate models, dashboards, or business decisions.