Abstract: The flaw
detection will detect data objects that are inconsistent with normal dataset.
In additional to normal data, there exist negative outliers in many
applications and data will be imperfectly labeled. This paper represents an
outlier detection approach to address data with imperfect data labels into
learning. Our past approach works in two steps. In the first step, we develop a
pseudo training dataset by computing possible values of each example based on
its local action. We introduce kernel k-means clustering method and
kernel Local Outlier Factor-based method to compute the likelihood values. In
the next step, we introduce the obtained possible values and limited abnormal
examples into SVDD-based learning to produce a more accurate classification for
global outlier detection. The proposed system has three approaches. They are
Naive Bayes approach, Logistic regression and . For classification of dataset
we go for these two approaches which makes easy to find outliers.
Keywords: Flaw detection, abnormal data, local outlier factor