Simplifying Logistic Regression

This week we will discuss Logic Regression! As always, we will avoid

technical details and focus on understanding the concept of algorithms and how

it is helpful in business. 

So, let’s get started.

A Logistic Regression in a Classification Model is used to predict what an

object is. The below examples provide different situations in which we would

use this model.

1. Wanted to predict if the fruit is an apple or an orange

2. If the car model is a Ford or Chevy

3. If email is or isn’t spam 

Data points like spam, car model, type of fruit are what we call Categorical

Data. This data is not numerical. The model takes a bunch of other data points

(aka variables) and uses them to determine the classification of the categorical

data point. There are many Logistic Regression models, and below, we break

them down.

1. Binary Logistic Regression: If the predicting outcome only has two cases.

Yes, or No, Spam or Not Spam, Black or White.

2. Multinomial Logistic Regression: We use this one if over two cases have no

order to them. Is the fruit and apple, orange or banana? 

3. Ordinal Logistic Regression: we use this process for multiple cases with an

order. Like predicting a restaurant rating of 1-5 or a movie rating.

This model is prevalent within the data community because it is easy to explain

its math. It is better to use a simple model that works well than a complicated

model that only works slightly better in the data world.  Another benefit of using this model is that it allows one to see the importance of each variable in predicting the categorical outcome. 

Let’s say you have the following variables to predict car model engine size,

mpg, weight, cost, and color. The logistic regression would allow you to see

precisely which variable used is the strongest indicator in determining the car’s

model.

In a business sense, this would be enormously helpful if you wanted to see how

critical certain variables are for customer retention, sale conversion, and movie

rating. These indicators provide the ability to understand your data and your

customers better. 

Data models are not as scary as everyone seems to think they are if we first

understand the overall concept and when it is appropriate to use them. There are

very few situations when you must calculate anything by hand. Remember, you

don’t need to understand how everything in your car engine works to drive your

car to the store. You only need to know how to turn it on, how to maneuver the

car’s apparatus, and the rules of the road.