Decision Tree Summary

Let's now look at some of Decision Tree's peculiarities.

1. Interpretability.
   Unlike most Machine Learning algorithms, Decision Trees are easy to visualize and interpret;
2. No data preparation required.
   The decision Tree requires none to very little data preparation. It does not need scaling/normalization. It can also handle missing values and does not suffer from outliers much;
3. Provides feature importances.
   While training, a Decision Tree calculates feature importances that represent how impactful each feature was to form the Tree. You can get feature importances using the .feature_importances_ attribute;
4. Computational complexity.
   Suppose m is a number of features and n is a number of training instances. The complexity of a Decision Tree training is O(n·m·log(m)), so the training is quite fast unless there is a large number of features. Also, the complexity of predicting is O(log(n)), so the predictions are fast;
5. Not suitable for large datasets.
   Although Decision Trees may work great for small sets, they usually don't work well for large datasets. Using Random Forest is preferable for large datasets;
6. Decision Trees are unstable.
   Small changes in hyperparameters or data may cause a very different tree. Although it is a disadvantage for a single Tree, it will benefit us in a Random Forest, as you will see in the next section.

And here is a little summary: