You are on page 1of 11

Logistic Regression

Ravi Mody
Content

▪ Qualitative Variable
▪ Classification
▪ Linear to logistic, log(odds).
▪ Fitting, Maximum Likelihood Method
▪ Goodness of fit
▪ Why log(loss)
▪ Coefficient interpretation
Qualitative variables

▪ What are qualitative variables?


▪ Dealing with qualitative explanatory variables.

▪ Dealing with qualitative response variable : Classification


Linear to Logistic (why?)

The logit distribution constrains


the estimated probabilities to lie
between 0 and 1
Maximum Likelihood & Logistic
Regression

• Best fit line: Using least square errors to find best fit line as in linear
regression is not possible, hence maximum likelihood (a method of
estimating the parameters of an assumed probability distribution,
given some observed data) is used.
Measuring goodness of fit: Deviance
Why SE cannot be used logistic regression?

▪ Hence it is very clear to us that LSE doesn’t strongly penalize


misclassifications even for the perfect mismatch!
Interpretation of Coefficient
Multinomial regression

You might also like