From AMS Glossary
A variation of multiple linear regression analysis for prediction of the occurrence or nonoccurrence of an event.
To account for the nonnumerical nature of the predictand, a discriminant function is used as a type of regression function usually derived in such a way that positive values of the function correspond to "occurrence" and negative values to "nonoccurrence." In meteorology, for example, the occurrence of precipitation (the predictand) can be related to measures of vertical velocity, dewpoint temperature, pressure change, and other variables (the predictors) through a discriminant function. Values of the function above or below a threshold value (typically, zero) can be used to predict precipitation occurrence.