In order to develop prediction instruments that have sufficient predictive power, it is essential to understand the specific domain the prediction instrument is developed for. This domain analysis is especially important for domains where human behavior, politics, or other soft factors play a role. If these are not well understood, the predictive power of the prediction instrument would be severely affected. In this paper, we provide literature based reasons for the use of domain analysis for the development of prediction instruments, and we discuss the circumstances under which domain analysis is especially important. We present a structured literature review of the actual adoption of domain analysis for predictive analytics. That shows that few papers discuss how domain analysis was performed, and when it is discussed, the type of analysis often does not fit with the type of domain. As these papers do show adequate predictive power, we believe that the domain analysis in these papers was done implicitly. To make the process of prediction instrument development, including domain analysis more transparent, we present requirements for a method for prediction instrument development , and an outline for such a method based on those requirements.
|Number of pages||10|
|Publication status||Published - 29 Mar 2016|
|Event||Big Data Interoperability for Enterprises (BDI4E) Workshop 2016 - Guimarães, Portugal|
Duration: 29 Mar 2016 → 30 Mar 2016
|Conference||Big Data Interoperability for Enterprises (BDI4E) Workshop 2016|
|Period||29/03/16 → 30/03/16|