Is Combining Classifiers with Stacking Better than Selecting the Best One?

By Saso Dzeroski, Bernard Zenko. 2004

936 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Ensemble-based classifiers (2010, methodology): “It has been shown that with stacking the ensemble performs (at best) comparably to selecting the best classifier from the ensemble by cross validation (Džeroski and Ženko 2004).”
Detection of malicious code by applying machine learning classifiers on static features: A state-of-the-art survey (2009, methodology): “It has been shown that with stacking the ensemble performs (at best) comparably to selecting the best classifier from the ensemble by cross-validation (Dzeroski and Zenko, 2004).”
Automatic Denoising of Functional MRI Data: Combining Independent Component Analysis and Hierarchical Fusion of Classifiers (2014, methodology): “Training the ensemble consists of extracting D , D sel , D a , D m , D a − sel and D m − sel , training the k-NN, decision tree and SVMs, on each of the datasets, and training the fusion-layer classi ﬁ er using the lower-level classi ﬁ ers' probabilistic outputs (Dzeroski and Zenko, 2004).”
A Machine Learning Framework for Stock Selection (2018, methodology): “As for Stacking, it’s risk proﬁle is somewhere between RF and DNN since it is a combination of RF and DNN.”
Activity Recognition from Accelerometer Data (2005, background): “Meta-level classifiers can be clustered into three frameworks: voting (used in bagging and boosting), stacking (Wolpert 1992; Dzeroski & Zenko 2004) and cascading (Gama & Brazdil 2000).”
Minority report in fraud detection: classification of skewed data (2004, methodology): “LIMITATIONS This paper should have used superior approaches such as SMOTE [9] to create the oversampled data partitions, and multiresponse model trees [16] as the meta-classifier.”
XSleepNet: Multi-View Sequential Model for Automatic Sleep Staging (2020, neutral): “This result is expected as combing multiple high-variance “models” (due to overfitting effect on small data), we are able to obtain an ensemble model with lower bias than any individual ones [54].”
Group Search Optimizer: An Optimization Algorithm Inspired by Animal Searching Behavior (2009, methodology): “Algorithm EENCL [62] EPNet [51] SVM-best [54] CCSS [55] OC1-best [56]”
Introduction to the Special Issue on Meta-Learning (2004, background): “The third contribution to this issue (Dzeroski & Zenko, 2004) is a study within the area of multiple classifiers (Section 2.2), in particular, within the learning paradigm known as stacking.”
MetaFraud: A Meta-Learning Framework for Detecting Financial Fraud (2012, methodology): “The training data for the top-level classifiers was generated by running the bottom-level classifiers using 10-fold cross-validation on the training instances (Dzeroski and Zenko 2004; Ting and Witten 1997).”
Evaluation of random forest and adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery (2008, methodology): “generated using the same training set, each from a different learning algorithm and their decisions are combined by simple voting, or more sophisticated methods like consensus theory (Benediktsson & Swain, 1992) and stacking (Džeroski & Ženko, 2004).”