The “gradient” in Gradient Boosting Machine is a reference to the concept of gradient descent.
XGBoost is a modern implementation of Gradient Boosting Machine that works largely the same as the standard GBM.
Adaboost (Adaptive Boosting) is a simple boosting technique that is a predecessor to modern algorithms like GBM and its offshoots.
A weak learner refers to a prediction mechanism that produces results that are only slightly more predictive than those resulting from a random chance model.
For any decision-tree based method, feature importance can be measured in a couple of ways.
Tuning the combination of number of trees and learning rate is a good way to ensure you are creating a model with appropriate complexity.
On structured datasets, a well-tuned GBM more often outperforms a Random Forest.
In Random Forest, decision trees are constructed independently and the results are aggregated through either averaging or majority vote after all trees are created.
Advantages: High accuracy
Disadvantages: Requires some computing power and time spent in parameter tuning
Key hyper-parameters for a GBM are: Number of trees, learning rate, and maximum depth