What is the Stacking Algorithm in Weka? How it actually is working?

Question

Is the result of Base classifiers are being selected by voting system & then what actually the Meta classifier is getting as it's input,whole classifier or just the miss-classified ones ? It would be helpful if the whole mechanism can be explained with a simple example like this link Majority vote algorithm in Weka.classifiers.meta.vote

Thanks in advance.

Answer 1

Consider an ensemble of n members. Each of these members are trained on a given set of training data. The ensemble members may share the same classifier type (homogeneous) or use different classifiers (heterogeneous). Diversity is encouraged between the members so that each member generates different estimates.

Instead of voting or averaging the estimates of each ensemble member to generate the ensemble estimate, stacking trains a meta-learner between the ensemble members and the ensemble estimate that accepts each ensemble member's estimate as inputs and generates the ensemble output.

An example is outlined below:

Ensemble Member 1 Estimate --->
Ensemble Member 2 Estimate --->  Meta-Learner -> Ensemble Estimate
Ensemble Member 3 Estimate --->

In the case of Weka, the 'classifiers' parameter refers to the ensemble members that are initially trained to estimate the output, and the 'metaClassifier' is the model that takes the classifier inputs and estimates the ensemble output.

Hope this helps!