Competing Risks in Healthcare and Predictive Maintenance
Competing risks occur when more than one event can happen, and one event changes or prevents the chance of observing another.
Read article
Measurement Error in Predictors: Regression Dilution and the Field Deployment Gap
The fitted effect of temperature comes out at half what the physics says. The model built on lab measurements loses two thirds of its accuracy on field senso...
Read article
Permutation Importance with Correlated Features: When the Ranking Lies
A near-duplicate sensor with no effect of its own outranks a feature that genuinely drives the outcome. Permutation importance is working exactly as designed...
Read article
When the Bootstrap Fails: Dependent Data, Small Samples, and Extremes
Resample the data, recompute the statistic, read off the spread. On an autocorrelated series of 200 points the interval that comes out is three times too nar...
Read articlePaired vs. Independent Samples: The Design Choice Behind the Test
The choice between paired and independent tests is not a software option. It is a statement about the study design and the dependence structure in the data.
Read article
Quantile Regression: Predicting the Range, Not the Average
The model predicts 36 minutes and 40 percent of deliveries take longer. The customer did not ask for the mean. They asked when the parcel would arrive, and t...
Read articleSlice-Based Model Evaluation: Finding the Failures Average Metrics Hide
Slice-based evaluation exposes where a machine learning model fails by breaking aggregate performance into meaningful subgroups, conditions, and operational ...
Read article
Data Drift and Fairness: Monitoring Equity When Populations Change
A fair model at launch can become unfair in production when populations, behavior, policies, or measurement systems change.
Read article
Equivalence Testing: Proving a Model Is No Worse
The cheaper model scores a non-significant p of 0.4 against the incumbent and is declared "no worse". With 50 test cases, a model that is truly 1.5 points wo...
Read article
Multiple Comparisons in Model Monitoring: Why the Alerts Never Stop
Test two hundred features every morning at the 5 percent level and you get about ten alerts a day with nothing wrong. After a week nobody reads them, and the...
Read article