Open AccessComparative Study of Outlier Detection Algorithms via Fundamental Analysis Variables
Author(s) -
Şenol Emir,
Hasan Di̇nçer,
Ümit Hacıoğlu,
Serhat Yuksel
Publication year - 2015
Publication title -
international journal of research in business and social science
Language(s) - English
Resource type - Journals
ISSN - 2147-4478
DOI - 10.20525/ijrbs.v4i4.462
Subject(s) - anomaly detection , outlier , computer science , data mining , novelty detection , local outlier factor , preprocessor , process (computing) , pattern recognition (psychology) , artificial intelligence , algorithm , novelty , philosophy , theology , operating system
In a data set, an outlier refers to a data point that is considerably different from the others. Detecting outliers provides useful application-specific insights and leads to choosing right prediction models. Outlier detection (also known as anomaly detection or novelty detection) has been studied in statistics and machine learning for a long time. It is an essential preprocessing step of data mining process. In this study, outlier detection step in the data mining process is applied for identifying the top 20 outlier firms. Three outlier detection algorithms are utilized using fundamental analysis variables of firms listed in Borsa Istanbul for the 2011-2014 period. The results of each algorithm are presented and compared. Findings show that 15 different firms are identified by three different outlier detection methods. KCHOL and SAHOL have the greatest number of appearances with 12 observations among these firms. By investigating the results, it is concluded that each of three algorithms makes different outlier firm lists due to differences in their approaches for outlier detection.