Evaluation of the Intelligence Collection and Analysis Process

Stark, Livia and Glazebrook, Kevin and Jacko, Peter and Atkinson, Michael (2022) Evaluation of the Intelligence Collection and Analysis Process. PhD thesis, Lancaster University.

Text (2022StarkPhD)
2022starkphd.pdf - Published Version
Available under License Creative Commons Attribution.
Download (8MB)

Abstract

Intelligence is a critical tool in modern security operations that provides insight into current and future operational conditions. It is a concept that transfers to other applications where monitoring activities or situations is imperative, such as ecological research. As technological advances in the past decades lead to increased availability of potential intelligence, we concentrate on source selection to ensure the resulting intelligence is of high quality and fit for purpose. We wish to bring focus to the more varied nature of intelligence than what is currently reflected in models of its collection and evaluation. Therefore, we examine the intelligence collection and analysis process in two separate scenarios; one treats it as a ongoing strategic activity, in another intelligence collection is carried out with an investigative intent. The first problem we formulate concerns source selection with a random time delay in feedback, corresponding to the collection and evaluation time of the intelligence. Both the distributions of such time delay and the outcome of the intelligence evaluation are unknown, giving rise to the classic exploration-exploitation dilemma in a long-run setting. We develop promising approaches to accommodate the novel features of the model based on Gittins indices and the knowledge gradient, and examine the issues presented when incorporating structures of dependence between the time delay and the outcome of the evaluation. Next, we develop a novel intelligence collection problem rooted in tactical level source selection, aiming to piece together an intelligence picture comprised of multiple types of information, for example, where and when an attack is planned. We demonstrate that when all elements of the model are known, dynamic programming provides the optimal policy. When some elements are unknown, which introduces an exploration-exploitation aspect to the model, we find that in certain cases the ability to learn is severely limited.