Bilkent University
Department of Computer Engineering

---

CS590/690 SEMINAR

 

Less is More: An Online Classifier Chain with Less Trainable Classifiers"

 

Onur Yıldırım
Master Student
(Supervisor: Prof. Dr.Fazlı Can)
Computer Engineering Department
Bilkent University

Abstract: Traditional data mining activities focus on static data, however, many companies generate massive amounts of data nowadays. Some data streams carry enormous amounts of information and instances arrive at high speed. The need for new algorithms and adaptation of existing algorithms arises in order to keep up with the speed, memory requirements, and changes in the data distribution. A great portion of the data stream classification research focuses on mining single-labeled data where each data instance is associated with only one label in its label set. Multi-label classification methods aim to classify data instances that are associated with one or more labels in the label set. We propose a hybrid model that benefits from the upsides of Label Powerset and Classifier Chains methods by enabling each classifier to train on multiple labels instead of one. As our approach employs less classifiers, it faster in most and accurate in some cases.

 

DATE: October 23th, Monday @ 15:50 EA-502