Designing dashboards for managing model lifecycle

Abstract

Data mining is being extensively used for analysis of large data collections. While there is previous work on dashboard support for visual mining of the data, there is little or no work on dashboard support for managing the lifecycle (e.g. health) of the data mining models themselves. Issues such as quick performance decay, large scale deployments, collaborative use, and real-time business integration of models necessitate this type of support. In this paper, based on a year long study, we first describe the six stages of the model lifecycle and the preliminary design of the backend system that helps users manage mining models. Next, we discuss the three dimensions to be considered for dashboard visualization of the model lifecycle: introspection, customization, and presentation.