Advanced Analytics – Paradigms, Tools, and Techniques - Getting Started with Greenplum for Big Data Analytics - page 78

Logistic regression
Logistic regression is the preferred method for many binary classification problems.
For example:
• True/false
• Approve/reject
• Buy/don't buy
Logistic regression is apt if we need the probability of an event against predicting
class variables. It is recommended that we try logistic regression in the first step for
all binary class problems. In a logical regression model, the outcome is determined
by a process like flipping a coin. The predictors that we know determine the process,
but the unknown determines the outcome here. Hence, we determine what change
in predictor changes the probability of the outcome.
Logisticregressionisalsocalledlogitmodel.Anexampleforlogisticregressionmod-
el is analyzing the factors that influence winning or losing in an election for a politi-
cian. The dependent variable would be binary, win, or lose and the predictor vari-
ables of interest can be the amount of money and time spent on the campaign,
demographic conditions of the candidate, and so on.
In a logistic function, as the response variable is binary, there is a curved relationship
seen in the plotting and this is referred to as sigmoid . The following figure shows
four variants of sigmoid curves (we can observe that the value for y oscillates
between 0 and 1):