ABSTRACT
Binary classification models or binary classifiers are predictive models whose response variable is binary, assuming only two possible values. Typically, these binary classification models predict not the response variable itself, but rather the probability of a positive response, where one of the two possible response values has been designated "positive". The basic predictive modeling problem considered here is that of predicting the compressive strength of laboratory concrete samples on the basis of sample characteristics, including the sample age in days and seven composition variables that describe the amounts of different components included in the concrete mix. An important issue in building and using predictive models is the impact of categorical predictors with thin levels. This chapter discusses the logistic regression model, which belongs to the larger class of generalized linear models that can be used to fit response variables that must be positive or that are discrete-valued count variables.
