Gradient Boosting

Tiya Vaj
2 min readFeb 9, 2024

Gradient boosting is a machine learning technique used for building predictive models, particularly for regression and classification tasks. In gradient boosting, the model is built sequentially by combining multiple weak learners, typically decision trees, to create a stronger predictive model.

Here’s what it means for gradient boosting:

  1. Sequential Model Building: In gradient boosting, weak learners, often decision trees, are added to the ensemble sequentially. Each new tree tries to correct the errors made by the previous trees, leading to a more accurate overall model.
  2. Gradient Descent Optimization: Gradient boosting minimizes a loss function by using gradient descent optimization. At each step of adding a new weak learner, the algorithm calculates the gradient of the loss function with respect to the current ensemble’s predictions. It then builds the new weak learner to minimize the error represented by this gradient.
  3. Gradient Descent Step: The gradient descent step in gradient boosting involves fitting a weak learner to the gradient of the loss function. This means that each new tree is trained to predict the residuals (the differences between the actual values and the current predictions) of the previous ensemble.
  4. Weighted Voting: In the ensemble, each weak learner’s contribution to the final prediction is weighted based on its performance in reducing the loss function during training. Weaker learners typically have less influence on the final prediction compared to stronger ones.
  5. Boosting: The term “boosting” in gradient boosting refers to the process of improving the model’s performance by iteratively adding new weak learners. Each new learner is focused on capturing the errors or patterns that the previous learners failed to capture, leading to a boosted overall performance.

Overall, gradient boosting is a powerful machine learning technique that builds predictive models by sequentially combining weak learners in a way that optimally reduces the error or loss function of the model.



Tiya Vaj

Ph.D. Research Scholar in Informatics and my passionate towards data-driven for social good.Let's connect here