Automatic Density Cluster Generation and Visualization Method Based on Overlapped Data Points

Abstract

Handling overlapped data points is a challenging task in data visualization because overlaps prevent identification of data points, especially, when the data density is very high. Removing, reducing, and reformulating are the most common techniques that used to overcome overlapping. Instead of removing, reducing, or reformulating overlap, a new method that uses overlap to generate colour-coded density clusters on a bitmap was proposed. The basic unit of a bitmap is a pixel which has n bits. Individual pixel is capable of holding 2n number of colours. Because each colour of a pixel can be converted in to a unique integer, a bitmap is a matrix of pixels as well as a matrix of integers. These integer values of a pixel can be mapped with values of a certain variable. Therefore, a pixel can be used as a single knowledge cell. In the present method, plotting area is a bitmap and the size and the shape of the marker (graphical symbol of a data point) will be selected in a manner that they can be overlapped. When there is an overlap, colour of such overlapped area can be added. If there is no overlap, keep the initial colour of the marker. Addition of the colour of overlapped area creates new colour that is different from original colours and colour value is proportional to the data density. This resembles an updating of knowledge. More overlapping (high density) make colour of the intersected area identical, even can be identified by a naked eye. Existing cluster analysing techniques need separate algorithm for finding clusters and visualizing technique for representing identified clusters. In contrast, the new method automatically creates clusters (without separate algorithm) by automatically generated colour lines which can be considered as isoclines. 35620 data points were mapped with a new method and the results showed very clear cluster formation over heat map and contour plot approaches