A0344
Title: Geometric goodness of fit measure to detect patterns in data point clouds
Authors: Alberto Hernandez - Universidad de Costa Rica (Costa Rica) [presenting]
Abstract: A geometric goodness-of-fit index is derived which is similar to $R^{2}$ using geometric data analysis techniques. We build the alpha shape complex from the data cloud projected onto each variable and estimate the area of the complex and its domain. We create an index that measures the difference in area between the alpha shape and the smallest squared window of observation containing the data. By applying ideas similar to those found in the closest neighbor distribution and empty space distribution functions, we can establish when the characterizing geometric features of the point set emerge. This allows for a more precise application for our index. We provide some examples with anomalous patterns to show how our algorithm performs.
