A Conformal Martingales Ensemble Approach for addressing Concept Drift

Charalambos Eliades, Harris Papadopoulos
Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications, PMLR 204:328-346, 2023.

Abstract

We propose an ensemble learning approach to tackle the problem of concept drift (CD) in data-stream classi cation. Accurately detecting the change point in the distribution is insufficient to ensure precise predictions, particularly when the selection of a representative training set is challenging or computationally expensive. More speci cally, we employ an ensemble of ten classi ers that use a majority voting mechanism to make predictions. To promote diversity among models, we train each on a different number of instances, resulting in different sequences of p-values and construct an Inductive Conformal Martingale (ICM) for each one. When the ICM algorithm detects a change point in the corresponding p-value sequence, we perform a retraining process of the corresponding classi er. We evaluate the performance of our proposed methodology on four benchmark datasets and compare it to existing methods in the literature. Our experimental results show that the proposed approach exhibits comparable and in some cases better accuracy than two state-of-the-art algorithms.

Cite this Paper

BibTeX
@InProceedings{pmlr-v204-eliades23a, title = {A Conformal Martingales Ensemble Approach for addressing Concept Drift}, author = {Eliades, Charalambos and Papadopoulos, Harris}, booktitle = {Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications}, pages = {328--346}, year = {2023}, editor = {Papadopoulos, Harris and Nguyen, Khuong An and Boström, Henrik and Carlsson, Lars}, volume = {204}, series = {Proceedings of Machine Learning Research}, month = {13--15 Sep}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v204/eliades23a/eliades23a.pdf}, url = {https://proceedings.mlr.press/v204/eliades23a.html}, abstract = {We propose an ensemble learning approach to tackle the problem of concept drift (CD) in data-stream classi cation. Accurately detecting the change point in the distribution is insufficient to ensure precise predictions, particularly when the selection of a representative training set is challenging or computationally expensive. More speci cally, we employ an ensemble of ten classi ers that use a majority voting mechanism to make predictions. To promote diversity among models, we train each on a different number of instances, resulting in different sequences of p-values and construct an Inductive Conformal Martingale (ICM) for each one. When the ICM algorithm detects a change point in the corresponding p-value sequence, we perform a retraining process of the corresponding classi er. We evaluate the performance of our proposed methodology on four benchmark datasets and compare it to existing methods in the literature. Our experimental results show that the proposed approach exhibits comparable and in some cases better accuracy than two state-of-the-art algorithms.} }
Endnote
%0 Conference Paper %T A Conformal Martingales Ensemble Approach for addressing Concept Drift %A Charalambos Eliades %A Harris Papadopoulos %B Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications %C Proceedings of Machine Learning Research %D 2023 %E Harris Papadopoulos %E Khuong An Nguyen %E Henrik Boström %E Lars Carlsson %F pmlr-v204-eliades23a %I PMLR %P 328--346 %U https://proceedings.mlr.press/v204/eliades23a.html %V 204 %X We propose an ensemble learning approach to tackle the problem of concept drift (CD) in data-stream classi cation. Accurately detecting the change point in the distribution is insufficient to ensure precise predictions, particularly when the selection of a representative training set is challenging or computationally expensive. More speci cally, we employ an ensemble of ten classi ers that use a majority voting mechanism to make predictions. To promote diversity among models, we train each on a different number of instances, resulting in different sequences of p-values and construct an Inductive Conformal Martingale (ICM) for each one. When the ICM algorithm detects a change point in the corresponding p-value sequence, we perform a retraining process of the corresponding classi er. We evaluate the performance of our proposed methodology on four benchmark datasets and compare it to existing methods in the literature. Our experimental results show that the proposed approach exhibits comparable and in some cases better accuracy than two state-of-the-art algorithms.
APA
Eliades, C. & Papadopoulos, H.. (2023). A Conformal Martingales Ensemble Approach for addressing Concept Drift. Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications, in Proceedings of Machine Learning Research 204:328-346 Available from https://proceedings.mlr.press/v204/eliades23a.html.

Related Material