Binary response modelling concepts

Learning objectives

• Interpret probability and odds in a binary model.
• Fit a simple logistic relationship.
• Assess discrimination and calibration cautiously.

Binary modelling basics

Logistic regression models log-odds as a linear function of X. Report odds ratios with CIs and predicted probabilities that operators can act on.

Balance, non-linearity, interactions and threshold choice all matter, especially with imbalanced classes.

Garment-factory example

Estimate probability of seam failure from SPI, fabric stretch, needle age and thread type; translate the fitted model to a probability curve technicians can read.

Method

1. Preserve unit-level binary data.
2. Fit the model in a free tool.
3. Convert coefficients to probabilities for communication.
4. Validate on new data; check calibration.

Common mistakes

• Treating odds ratio as risk ratio.
• Choosing a threshold for maximum accuracy in imbalanced data.