A BRIEF SURVEY ON CLASSIFICATION METHODS FOR UNBALANCED DATASETS

Manoj Kumar Sahu, Rajeev Pandey, Sanjay Silakari

Abstract: In real world, we deal with the data sets which are unbalanced in nature. Information sets are lopsided when no less than one class is spoken to by extensive number of preparing illustration (called greater part class) while different classes make up the minority. Due to this uneven nature of information sets we have great precision on dominant part class yet on the other side exceptionally poor exactness on the minority class, while we attempt to foresee the class enrollment. Accordingly, the lopsided way of information sets can have a negative impact on arrangement execution of machine learning calculations. Specialists have been made numerous endeavors to manage such issues of order of information at information level and additionally calculation level. In this paper we speak to a brief study of existing answers for the class-unevenness issue proposed both at the information and algorithmic levels.

Keywords: Classification, Cost Sensitive Learning, Unbalanced Data, Over Sampling, Under Sampling.

DOI: https://doi.org/10.15623/ijret.2016.0507023