Necessity of all three base classifiers in TIGER

Determine whether TIGER requires all three base classifiers—Ridge, Extra Trees, and Naive Bayes—or whether a reduced portfolio, such as Ridge and Extra Trees alone, can match TIGER’s accuracy at lower computational cost.

Background

TIGER applies Ridge, Extra Trees, and Naive Bayes to each of four time-series representation families, yielding twelve base learners. Although this shared portfolio contributes to the stacking ensemble, the paper does not establish whether every classifier is necessary. Removing one or more classifiers could reduce computational cost, but the effect on predictive accuracy remains unresolved.

References

Fourth, we have not investigated whether all three base classifiers in TIGER's shared portfolio (Ridge, Extra Trees, and Naive Bayes) are necessary. A reduced portfolio, for instance Ridge and Extra Trees alone, might match TIGER's accuracy at a lower computational cost. We leave this investigation to future work.

— TIGER: Time-Series Classification with In-Context-Learning Gated Ensemble of Representations  (2610.06156 - Faouzi, 5 Oct 2026) in Section 7, fourth limitation