Detection of Outliers through Influence Function on Affinity

Abstract

Outliers are the atypical observations that lie at abnormal distances from the other observations in a random sample. Such outliers are often seen as contaminating the data. In general, the rejection of influential outliers improves the accuracy of the estimators and so the results with the identification of outliers have become the most important aspect in any data analysis. Outlier detection finds many applications in the areas such as data cleaning, fraud detection, network intrusion, pharmaceutical research and exploration in science data buses. The distance based outlier detection is the most commonly used method. In this paper, the influence function for affinity is explained and the detection of outliers in classification problems using influence function for affinity is illustrated for univariate data through a few examples.