Outlier Detection

Most existing research in finding outlying aspects of an object, has been dealing with a static large data set. By querying a specific object against the data set, we will always get the same outcome. Unless the data set is manipulated in one way or another. In this proposal, we approach the problem of mining outlying aspects from a different perspective. Instead of using a constant data set, we use a real-time environment for an ever-growing and expanding data set. Where we attempt to find the outlying aspects of a query, as data is constantly increasing and changing. This adds another layer of complexity to finding outlying aspects, not only is it computationally heavier with the addition of real-time data. But the aspects that make an object special could be constantly changing and at one point in time, the object could be considered an inlier. Then after a few minutes of streaming data, could be an outlier in respect to certain aspects. We propose the adoption of widely used methods (Local …show more content…
Outlying data points are generally found through two different approaches either being a distance (Knorr & Ng, 1998; Knorr & Ng, 1999; Ramaswamy et al, 2000) or density method (Jin et al, 2001; Papadimitriou et al, 2003). These techniques find outliers based on the entire space of the data set. But in the previous few years the stage has shifted.
We are not concerned with the data points that deviate in the entire space. But instead we are focusing on how a data point becomes an outlier or special when looking at it through different dimensions . A great deal of effort is being invested into the research of outlying subspaces due to the fact that, as a data set increases with a high set of dimensions. It becomes computationally expensive and time consuming to