A Critical Review on Concept Drift Monitoring Process for Class Imbalance in Data Streams

Nouhaila Aasoum, Ismail Jellouli, Souad Amjad

2021

Abstract

Machine learning techniques have participated in world evolution. They have accomplished worthy goals in many areas such as banking, industry, cybersécurité, and many others. However, in most data analysis applications, data comes in streams based on online learning scenarios. As streams emerge and change quickly over time, it will be hard to store them in memory. Thus, the analysis has become a real challenge to mitigate using traditional approaches. The change in data distribution degrades the accuracy performance of the trained model and becomes inefficient. This phenomenon is called concept drift, where the model must adapt quickly to these changes, including those in the environment, trends, or behaviour, to maintain their accuracy. Another phenomenon commonly exists in real-world applications is a class imbalance, when data distribution changes within classes. Thus, the model will favor the majority class and ignores the minority one. The problem becomes more challenging when both of them co-exist. Therefore, a few studies addressed this research gap. The objective is to detect concept drift with class imbalance for enhanced performance in a non-stationary data environment. This study will focus on class imbalance handling techniques, and concept drift effects in decreasing model performance, and some methods to detect concept drift while existing class imbalance issue.

Paper Citation

in Harvard Style

Aasoum N., Jellouli I. and Amjad S. (2021). A Critical Review on Concept Drift Monitoring Process for Class Imbalance in Data Streams. In Proceedings of the 2nd International Conference on Big Data, Modelling and Machine Learning - Volume 1: BML, ISBN 978-989-758-559-3, pages 404-408. DOI: 10.5220/0010735500003101

in Bibtex Style

@conference{bml21,
author={Nouhaila Aasoum and Ismail Jellouli and Souad Amjad},
title={A Critical Review on Concept Drift Monitoring Process for Class Imbalance in Data Streams},
booktitle={Proceedings of the 2nd International Conference on Big Data, Modelling and Machine Learning - Volume 1: BML,},
year={2021},
pages={404-408},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010735500003101},
isbn={978-989-758-559-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Big Data, Modelling and Machine Learning - Volume 1: BML,
TI - A Critical Review on Concept Drift Monitoring Process for Class Imbalance in Data Streams
SN - 978-989-758-559-3
AU - Aasoum N.
AU - Jellouli I.
AU - Amjad S.
PY - 2021
SP - 404
EP - 408
DO - 10.5220/0010735500003101