Egg Shape as a Generalized Radial Basis Shape for Discard-After-Learn Method in Streaming Data

Various density-based learning methods have proposed learning algorithms which use symmetrical geometric shapes to represent the decision boundaries such as circle, ellipse. These shapes are incapable of handling non-symmetrical data distribution such as skewed distribution. To cope with this type of distribution, we proposed a new structure of the data capsule derived from Yamamoto's equation of egg shaped curve along with a new evaluation metric for measuring the density of the data capsule. This egg equation is a generic shape equation which can be parametrically transformed into the equations of ellipse and circle, which makes it feasible to represent both symmetrical and non- symmetrical data distributions. The experiment showed that the proposed method gives the highest density when the distribution has a degree of skewness and still gives similar result when the distribution of the data sets is symmetric compared to traditional circular and elliptical data capsules.