Outlier detection from multiple data sources

The preparation of a dataset by merging multiple data sources using the fusion method may lead to loss vital information from each multi-source and certain amount correlative among sources. Based on extensive analysis “the unique characteristics” outliers, we propose outlier detection techniques reliably identify outliers in datasets. Several real-world examples are considered classify into three types (Type I-III) depending correlation We design baseline algorithm, which is an intuitive solution, optimal algorithm known as multiple-data-sources oriented (MOD) obtain high-score outliers. In addition, build MOD+ speed up process. A new density metric combining kNN RNN introduced evaluate deviation degrees applied develop outlier-join operators. MOD adept at (1) mining one datasets (2) sensing these implement algorithms synthetic experimental results demonstrate that proposed methods promising practical context detecting