Detection of the Outliers for Large Scale Data

Prapulla C.; Prof. Malatesh S. H.

In data mining and machine learning the detection of the outliers has become a very important topic. An effective and efficient framework is needed to identify deviated data in many real world applications such as intrusion detection and credit card fraud. Many methods used for detection of the outliers are typically implemented in batch mode and that implementation on large scale data is difficult. For implementation of the batch mode on large scale data will lead to sacrifice of computation and memory requirement. In this paper, I address the computational and memory management issues and propose Online oversampling principal component analysis algorithm that aims at detecting outliers from large scale data via online updating technique. The prior principal component analysis based approaches, the data would not be stored in covariance matrix, and thus our approach would be basically interested in large scale data problems. In this algorithm the oversampling of target instances and extracting the principal direction of the data the algorithm allows us to determine the target instances according to the variation of the resulting eigen vector. The proposed framework need not have to explicitly compute the eigen vector and hence this favours the problem that is being addressed in the paper. The eigen vector is not computed explicitly and hence the limitation of computation and memory management is favoured. On comparison of other methods the proposed experimental method will be both accurate and efficient.

Select Volume / Issues:

• Volume 5,Issue 3,May 2014

Year:

2014

Type of Publication:

Article

Keywords:

Data Mining; Outlier Detection; Eigen Vector; Computational Requirement; Memory Management; Large Scale Data

Journal:

IJECCE

Volume:

5

Number:

3

Pages:

485-489

Month:

May