Bilkent University
Department of Computer Engineering

---

CS 590/690 SEMINAR

 

A novel ensemble approach based on Maximal Marginal Relevance (MMR) for Drifting Stream Classification

 

Soheil Abadifard

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

Abstract: Classification models must handle concept drift efficiently and effectively in a data stream environment. Ensemble methods are widely used for this purpose; however, the ones available in this area either update the model with a large data chunk or learn the data one by one. In the former case, the model may miss changes in the data distribution, while in the latter case, the model may be inefficient and unstable. To address these issues, we propose a novel ensemble approach based on Maximal Marginal Relevance (MMR), in which mini chunks are used at each update, and base classifiers are chosen by taking both accuracy and diversity into account at the same time. MMR is a method that was developed to be used in Information Retrieval (IR) systems.Regarding IR systems, this method reduces redundancy while maintaining query relevance when re-ranking retrieved documents. In terms of ensemble classification on stream data, we try to select classifiers with the highest accuracy while being diverse to maximize the overall system's accuracy. Our proposed approach employs a novel selection method that improves overall system accuracy and is time efficient. We will conduct extensive experiments with 20 datasets to demonstrate our model's adaptability to various types of drift.

 

DATE: April 3, Monday @ 15:30 Place: Zoom