I have a dataset whose response is binary(Yes/No) and the predictors are categorical(Department, Job role, marital status) as well as continuous (Salary, hours of work). So, can we use LDA for such dataset? Will similar logic apply to KNN?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ LDA is a particular form of Canonical correlation analysis (CCA). When there is a mixed type variables, Categorical CCA is appropriate. Its subtype when there is one dependent variable vs independents ia known as Categorical Regression. All these procedures are based on optimal scaling transforming the problem to classic linear OLS. Other, intrinsically, nonlinear procedures, include logistic regression and other Generalized linear models. $\endgroup$– ttnphnsDec 3, 2017 at 5:39
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Linear Discriminant analysis will work for the categorical data as long you can convert it to some numerical component.
There are assumptions of normality which may not be true for your data and you should care more about that.
