While Decision Trees are easy to understand, they have a major weakness: they tend to overfit the data. In this post, we’ll see how Random Forests solve this problem.
The Problem with Single Trees 🌳
Decision Trees work by drawing “boxes” around data. If a tree is allowed to grow too deep, it will create tiny, specific boxes for every single data point in your training set. This is called overfitting. The tree becomes great at “remembering” the training data but terrible at “predicting” new data.
Pruning
One way to fix this is Pruning—cutting back branches that don’t add much predictive power. This helps, but it isn’t always enough.
The Solution: Random Forests 🌳🌳🌳
A Random Forest is an “Ensemble” method. Instead of relying on one tree, it builds a whole forest of them.
How it works:
- Random Sampling (Bagging): Each tree in the forest is trained on a random subset of the data.
- Feature Randomness: Each time a tree needs to split, it only considers a random selection of the available features.
- Voting: When it’s time to make a prediction, every tree in the forest “votes.” The forest then picks the majority result (for classification) or the average (for regression).
By randomizing the data and the features, the trees become different from one another. While one tree might overfit, the average of 100 different trees will be much more stable and accurate.
Why use Random Forests?
- Highly Accurate: One of the most powerful algorithms for structured data.
- Robust to Overfitting: Much better than a single decision tree.
- Handles Missing Data: Works well even if some data is missing.
Real-World Uses
- Banking: Detecting loyal vs. fraudulent customers.
- Medicine: Identifying the correct combination of components in medicine.
- E-commerce: Predicting whether a customer will recommend a product.
Exercise
- Research how many trees are usually in a “typical” Random Forest.
- Compare a Random Forest implementation in Python (using
scikit-learn) vs. other languages. You’ll see why Python is the king of ML!
Next up, we dive into the “geometric” world of Support Vector Machines (SVM).
Written by
Abdur-Rahmaan Janhangeer
Chef
Python author of 9+ years having worked for Python companies around the world
Suggested Posts
Machine Learning Part 6: Entropy and Information Gain
In the previous post, we saw how Decision Trees split data based on “purity.” But how do we measure ...
Machine Learning Part 5: Decision Trees and Mixed Methods
Some machine learning methods are versatile enough to be used for both Classification and Regression...
AdaBoost vs. Naive Bayes Algorithms in Machine Learning
In machine learning, the algorithm plays a significant role while training and building a successful...
