Decision Tree Advantages
Decision Tree is an algorithm that caters a niche need in machine learning ecosystem. If you need to understand, observe, demonstrate and interpret not just the data but how exactly machine learning algorithm performs on it at each step than decision tree will be a perfect fit because that’s exactly what it’s capable of doing and competition is not.
If you need to increase robustness or inference accuracy you can always resort to ensembling decision trees which gives random forests.
Here are the main advantages of decision trees and what they might mean for your project.
1) High Interpretability
Decision Trees are highly interpretable machine learning models.
You can observe the data prediction relationships as well as inner workings of the model which can often be a very useful option.
If your project involves education, presentation or wrapping your head around the dataset this can be your model.
2) Fast Training
Decision Trees are relatively fast models and they scale well with efficient computational complexity.
If runtime and resources are particular restrains decision trees can offer unique versatility without the expense of resources.
3) Easy Data Prep
A strong selling point of decision trees is that it’s not a picky model when it comes to data.
Decision trees will handle missing values, will perform well with noisy data and won’t mind if data is linear or polynomial or completely non-linear in nature.
This means easy to no data prep, something most data scientists and analysts seem to enjoy dearly.
Many Options for Customization
Benefits of Kerneling
With the SVMs you see the results of optimizations tremendously and predictions with very high accuracy can be achieved this way. This also supports learning more about the data.
Are decision trees capable of accurate predictions?
Are decision trees accurate models? Well, sort of. Decision Trees are capable of making highly accurate predictions. In fact, in its heyday before deep learning was popular decision trees were the go-to algorithm for complex problems such as the ones that exist in computer vision domain.
Decision trees didn’t only offer accuracy and computational efficiency but also interpretability meaning, one could observe each step of the inner working of the algorithm and see what steps were taken while reaching a conclusive classification or regression result.
However, since Neural Networks and especially Deep Learning algorithms almost doubled the prediction accuracy of decision trees, it can be said that they lost some competitive edge. Also, developments in the hardware enabled historically more complex and slower algorithms to become feasible alternatives in machine learning tool-set.
Decision trees mostly live through random forests in most practical applications but they still find room for their interpretability edge in quite a few cases.