Evolutionary Online Machine Learning from Imbalanced Data

Abstract

The discipline of machine learning has raised plenty of well-understood and partially well-studied challenges. Research has been concerned with issues such as incompletely labeled or missing data, dataset imbalances regarding the distributions of the target values, as well as the non-deterministic and unpredictable behavior of non-stationary environments. In this article, one particular challenge will be reviewed and motivated - the challenge of online learning from imbalanced data common in real world environments. It is hypothesized how interpolation between already gained knowledge and a proactive exploration of the input space may lead to beneficial effects when learning from data streams exhibiting imbalances. After the definition of this doctoral study's objectives, a reference evolutionary online machine learning technique is briefly introduced. On this basis, all aspects that will be thoroughly investigated are sketched and finally integrated into a research schedule.