Skip to content

<<< Random Forest Algorithm

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

Random Forest Advantages

Random Trees offer the best of both worlds. Reliability, simplicity and low maintenance of decision trees, increased accuracy, decreased feature reliance and better generalization that comes from ensembling techniques.

1- High Accuracy

Random Forest is a very convenient algorithm that can deliver highly accurate predictions even out of the box.

Since it’s an ensemble algorithm, training multiple decision trees offers many benefits. Even a random forest with a single tree will usually outperform a decision tree model.

Randomness of samples and randomness of features also mean learning with less bias through increased variance during training.

 

2- Less Overfitting

Although overfitting is a problem with Random Forests like any other ML algorithm, they tend to overfit less and still be a lot more accurate than competition with the right settings, which are easily obtained.

One of the reasons why overfitting is a nuissance is because it can be hard to detect, but at least in this case it’s not so hard to fix.

3- Fast to Predict

Inference phase with Random Forests is fast. You can work with big data or make real-time Random Forest deployments without having performance problems.

One thing to note is if data is too big and dimensional at the same time Random Forest can be heavy on the memory aspect of the computation which can be addressed by optimizing feature related parameters.

Stable and Robust

4- Stability Related Random Forest Benefits

This is a main contributor to why people absolutely why Random Forest algorithm. We will list data related pros of Random Forests which will likely make you a fan also.

Data Champ

5- Data Related Random Forest Benefits

This is a main contributor to why people absolutely why Random Forest algorithm. We will list data related pros of Random Forests which will likely make you a fan also.

a) Immune to noise, bias and outliers

Random Forests just like Decision Trees do  a great job with missing data and natively handles it. 

b) Handling Categorical as well as Numerical Data

Random Forests just like Decision Trees do  a great job with missing data and natively handles it. 

c) Handling Missing Data

Random Forests just like Decision Trees do  a great job with missing data and natively handles it. 

d) No Scaling Needed

Since Random Forest is based on trees and trees don’t care about the scales of input Decision Trees as well as Random Forests are natively invariant to scaling of inputs. With any other machine learning algorithm that relies on some sort of distance calculation such as Support Vector Machines, Linear Models and kNN this won’t be the case. A welcome feature indeed if you’re not keen on scaling transformations.

Ask any seasoned Data Science practitioner and they will tell you Data Science is 80% to 90% data wrangling and 10% to 20% Machine Learning and AI. They will also often add how they don’t like dealing with data prep.

Well we believe you should resists the urge to follow this herd instinct and embrace data preparation processes because it’s just a reality and huge part of the Machine Learning and Data Science domains.

However, it does help to have an algorithm like Random Forest in the toolbox to just handle whatever data you throw at it like a champ.

Random Forest is also fast

6- Performance Related Random Forest Benefits

Random Forest is very fast when it comes to prediction. Its inference phase is very fast and training phase is usually fast enough and can be easily tuned to be faster.

Random Forest is cool

Summary

Summary is Random Forest continues to impress industry practitioners and academia alike. It has so much to offer under so many different conditions. It’s not a picky algorithm when it comes to dataset characteristics.

Random Forest has almost no significant shortcoming that will hinder a typical machine learning project. Although I’m sure there can be unique cases that will opt-out of using Random Forest algorithm for a specific reason. Random Forests still do have some disadvantages but these are light ones and can be easily addressed through tuning.

Recommended Articles: