Skip to content

<<< Logistic Regression Algorithm

  1. Overview
  2. Classification
  3. Regression
  4. Advantages
  5. Disadvantages
  6. Logistic Regression Complexity
  7. Tuning Logistic Regression
  8. Who Invented Logistic Regression?
  9. Logistic Regression Example 1

Logistic Regression Disadvantages

In this article we will take a look at common disadvantages of Logistic Regression that can turn analysts or data scientists off.

Although Logistic Regression is an algorithm with a strong fan base there are definitely one or two cons that render it less useful or more tricky to work with. We tried to compile a fair list of some of those disadvantages below.

1- Limited Application

Logistic Regression can be a picky machine learning model and that limits the use cases greatly.

First of all, Logistic Regression can only solve classification problems. So it won’t be able to predict continuous values.

Additionally, Logistic Regression works well with linearly separable datasets but non-linearity is not its strong suit either.

 

2- Overfitting

Overfitting is a common problem in Machine Learning world. Basically it means data learns training samples too closely and it’s not able to perform very well with different datasets during inference.

This is a valid concern with Logistic Regression as well however, this algorithm is loaded with regularization hyperparameters that are ready to be tuned which is a welcome option.

Tuning Logistic Regression

3- No Complexity

In addition to not being able to handle non-linearity, Logistic Regression is also not very good at dealing with outliers, other complex data relations or nuances.

This can be a problem if project requires uncovering sophisticated insights from the data.

Aside of that missing data will cause Logistic Regression to perform badly. Missing values will have to be taken care of adding another crucial extra step to the data pre-processing tasks.

Data preprocessing steps

4- Scaling

Scaling features is a very important step for Logistic Regression or it might make inaccurate predictions. Some people prefer as little data pre-processing steps as possible and this can be an unattractive necessity causing extra steps during model implementation. Although once you get used to the scaling steps it can be pretty practical and quick. Luckily Scikit-Learn offers great tools to take care of scaling and normalization.

If you still prefer a robust, accurate and efficient model that doesn’t require scaling you can check out random forests below:

Cons of Logistic Regression

Summary

In this article we have listed the commonly observed disadvantages of Logistic Regression algorithm. Despite its shortcomings Logistic Regression is a unique algorithm because it can output probabilities and its a statistical model which some people prefer to work with.

Also Logistic Regression has a very old history and has been very useful for many decades or even centuries. You can read a post about Logistic Regression history below: