Department of Computer Engineering
CS 590 SEMINAR
Ensemble Learning for Multi-label Classification in Concept Drifting Data Streams
Computer Engineering Department
As data streams become more prevalent, the necessity for online algorithms that mine this transient and dynamic data becomes clearer. Multi-label stream classification (MLSC) is a supervised learning problem where each instance in the data stream is classified into a subset of labels rather than just one of them. Many proposed methods in the literature on MLSC resort to ensembling strategies in order to increase their predictive performance and address changes in the data distribution over time (called concept drifts). We introduce an online evolving stacked ensemble with dynamic weighting, called GOOWE-ML, that utilizes spatial modeling to assign optimal weights to its component classifiers. We generate 4 models of GOOWE-ML using different base multi-label methods and compare them with 7 baseline models in the literature over 7 datasets. We validate our results in terms of statistical significance using Friedman test with Nemenyi post-hoc analysis. Empirical evidence shows that GOOWE-ML yields consistently better predictive performance (and lower average ranks) in almost every dataset with respect to the other prominent ensemble models.
DATE: 12 November, 2018, Monday, CS590 presentations begin at @ 15:40