Visual Data Mining Methods for Kernel Smoothed Estimates of Cox Processes

Abstract

Real world planning of complex logistical organisations such as the fire service is a complex task requiring synthesis of many different computational techniques, from artificial intelligence and statistical or machine learning to geographical information systems and visualization. A particularly promising approach is to apply established data mining techniques in order to produce a model and make forecasts. The nature of the forecast can then be rendered using visualization techniques in order to assess operational decisions, simultaneously benefiting from generic and powerful data mining techniques, and using visualization to understand these results in the context of the actual problem of interest which may be very specific. Previous approaches to visualization in similar contexts use iso surfaces to visualize densities, these methods ignore recent improvements in interactive 3D visualization such as volume rendering and cut-planes, these methods also ignore what is often a key problem of interest comparing two different stochastic processes, finally previous methods have not paid sufficient attention to differences between estimation of densities and point processes (or Cox processes). This paper seeks to address all of these shortcomings and make recommendations for the trade-offs between visualization techniques for operational decision making. Finally we also demonstrate the ability to include interactive 3D plots within a paper by rendering an iso surface using 3D portable document format (PDF).