Introductio to Multivariate Logistic Regression - Model Building

Welcome to the session on 'Multivariate Logistic Regression (Model Building)'.

Just like when you’re building a model using linear regression, one independent variable might not be enough to capture all the uncertainties of the target variable in logistic regression as well. So in order to make good and accurate predictions, you need multiple variables and that is what we’ll study in this session.

Before starting with multivariate logistic regression, the first question that arises is, “Do you need any extensions while moving from univariate to multivariate logistic regression?” Recall the equation used in the case of univariate logistic regression was:

$P=\frac{1}{1+e^{-({\beta }_0+{\beta }_{1}X)}}$

The above equation has only one feature variable $X$, for which the coefficient is ${\beta }_1$. Now, if you have multiple features, say n, you can simply extend this equation with ‘n’ feature variables and ‘n’ corresponding coefficients such that the equation now becomes:

$P=\frac{1}{1+e^{-({\beta }_0+{\beta }_1{X}_1+{\beta }_2{X}_2+{\beta }_3{X}_3+...+{\beta }_n{X}_n)}}$

Recall this extension is similar to what you did while moving from simple to multiple linear regression.

In this session

In this session, you will learn how to:

• Build a multivariate logistic regression model in Python
• Conduct feature selection for logistic regression using:
  • Automated methods: RFE -Recursive Feature Elimination
  • Manual methods: VIF and p-value check

We will use the ‘Telecom Churn’ dataset in this session to build a model using multivariate logistic regression. This will involve all the familiar steps such as:

• Data cleaning and preparation
• Preprocessing steps
• Test-train split
• Feature scaling
• Model Building using RFE, p-values and VIFs

Apart from the familiar old steps, you’ll also be introduced to something known as a confusion matrix and you’ll also learn how the accuracy is measured for a logistic regression model.
 

Prerequisites

There are no prerequisites for this session other than the knowledge of the previous session and the previous module.