Question

A data scientist at an insurance company needs to build a model that outputs the probability that a customer will file a claim within the next year. The target is a binary label (claim / no claim), but downstream business rules require a calibrated probability between 0 and 1 so that premiums can be tiered. A colleague suggests using linear regression on the 0/1 label. Which modeling approach is most appropriate for producing valid probability estimates for this binary outcome?

Accepted Answer

Use logistic regression, which applies a sigmoid function to produce outputs bounded between 0 and 1 suitable as probabilities

Answer

Use linear regression on the 0/1 label and clip any predictions outside the [0, 1] range

Answer

Use k-means clustering to group customers, then assign the cluster mean as the probability

Answer

Use a standard linear regression and interpret the raw continuous output directly as a probability