Troika – An improved stacking schema for classification tasks

Improving malware detection by applying multi-inducer ensemble

“…In this paper we introduce three main contributions: (1) we show that multi-inducer ensembles are capable to detect malwares; (2) we introduce an innovative combining method, called Troika [Menahem, 2008] which extends Stacking and show it superiority in the malware detection tasks; and (3) we present empirical results from an extensive real world study of various malwares using different types of features.…”

Taxonomy for characterizing ensemble methods in classification tasks: A review and annotated bibliography

“…Stacking is usually employed to combine models built by different inducers. There are several extensions to the stacking approach, including StackingC [150], Troika [107] and SCANN (Stacking, Correspondence Analysis and Nearest Neighbor) [110].…”

Impact of an Extra Layer on the Stacking Algorithm for Classification Problems

“…To solve this problem, Seewald (2002) Stacking efficiency is directly dependent on the number of classes of the problem (Jurek et al, 2014). A new approach called Troika was proposed by Menahem et al (2009) to address multi-class problems. It is based on the four-layer architecture, where the last layer contained only one model: The super classifier, that outputs a vector of probabilities as a final decision of ensemble.…”

“…The Stacking method offers certain benefits compared to Bagging and Boosting, including the ability to combine different classifiers with simplicity and having a final performance similar to the best classifier of the committee (Menahem et al, 2009). However, in multi-class problems, Stacking may perform worse than other meta-approaches.…”