Backward Elimination in Multiple Linear Regression | Read Now

When constructing any Machine Learning model, Backward Elimination is a selecting features strategy. It’s employed to get rid of characteristics that wouldn’t have much of an influence on the target variables or output projection. In the ML domain, there are several approaches to develop a classifier, including:

  1. All in methodology
  2. Back-ward elimination methodology
  3. Forward selection methodology
  4. Elimination in birecdirections
  5. Comparing the scores

The methodologies for building a framework in ML are listed above, however, we will simply employ the Backward Elimination procedure because it is the quickest.

How can Backward Elimination be applied to Multiple Linear Regression?

The procedures of backward elimination are as regards:

Step-1: To remain in the model, just choose the level of significance (e.g., SL = 0.07).

Step-2: All potential predictors should have been included in the framework.

Step-3: Take the predictor with the largest P-value into evaluation. Go to location d if P>SL.

Step-4: Getaway or eliminate the predictor.

Step-5: Replace this variable of this research and repeat step 3 until the condition is false.

Why do we need this method?

  • In an earlier Multiple Linear Regression blog, we explored a difficulty that is connected to this notion.
  • The only flaw in that modeling was that it’s not ideal because we had considered all of the independent factors and had no means of knowing which independent variable was most and lowest effecting the forecast.
  • Inappropriate features add to the model’s complexities.
  • As a conclusion, it is preferable to only have the main key features and keep our system short in order to obtain the best performance.
  • Hence, in hopes of improving the effectiveness of the algorithm, we’ll apply the Backward Elimination procedure.
  • This procedure is employed to optimize the MLR’s quality of the model by only including most prominent factors and excluding the lowest significant ones.
  • Backward elimination encourages the design to be fitted to the best scenario.
  • As a response, adopting backward elimination in a modeling is advised.


  1. Rapid Training: If unnecessary features are excluded from the framework, the system is trained with a list of alternative pattern features in a very small time. This rapid training comes under the scenario o nly when the system is working with major features and eliminating all noisy variables.
  2. Minmal model complezity: In attempt to lessen the model’s difficulty, the backward elimination process is employed to layoff undesirable characteristics from the model. It optimizes the model’s functionality. Only a few essential characteristics are needed to draw a decent fit with reasonable precision.
  3. Zero over-fitting: The Backward Elimination procedure produces zero over-fitting as it reduces the amount of features by clearing out the unproductive ones.


  1. In the backward elimination methodology, it is impossible to determine which predicator is to responsible for a further predicator’s refusal because of its nothingness.
  2. Any characteristic that has been eliminated from a prototype employing a Backward Elimination technique cannot be chosen again.
  3. The model’s rules for choosing the significance level are strict.

End Notes

  • To boost the model’s effectiveness and decrease its complication, this methodology is centrally employed. 
  • It is commonly utilized in regression analysis where the system must manage with a huge dataset.
  • It is a straightforward and clear strategy, as contrasted with forwarding selection and crossed validations, which lead in an excess of optimizations.
  • The Backward Elimination methodology starts by removing features with a larger significance level.
  • Its main purpose is to improve the system and prevent over-fitting conditions.

Leave a Reply

Your email address will not be published. Required fields are marked *