Diminishing Performance by Imbalanced Data in ML


The Machine Learning (ML) applications are an inseparable part of many research activities if not all. ML extracts information that is beyond human capacity. Thus, it identifies hidden relations in the dataset or makes predictions for the future. Also, the improvements in ML are an open research question. Between many aspects, imbalanced datasets reduce the ML performances significantly. ML algorithm predicts for the favor of the majority class. There are many attempts to overcome this problem. The most common method is an improvement on the dataset such as reducing the majority class or increasing minority class. In this presentation, we will explain the problem in detail and our recent attempts with the applications.


Dr. Serkan Güldal is a fulltime faculty in Physics Department at Adiyaman University. Dr. Güldal received his B.S. in Physics from Erciyes University in 2006. He obtained his master’s degree in 2011 in the field of Physics from Mississippi State University. In his master's study, he studied molecular dynamics by Density Functional Theory.


His Ph.D. is in Interdisciplinary Engineering from the University of Alabama at Birmingham (UAB) in 2016. His research indicates the redefinition of the least action principle outside mechanical physics. Since he obtained his Ph.D., he is interested in developing new algorithms. Currently, he is participating in USEPE project as a machine learning expert in Kongsberg, Norway.