Optimizing classification in intelligence processing

Abstract

The intelligence making process, often described as the intelligence cycle, consists of phases. Congestion may be experienced in phases that require time consuming tasks such as translation, processing and analysis. To ameliorate the performance of those timeconsuming phases, a preliminary classification of intelligence items regarding their relevance and value to an intelligence request is performed. This classification is subject to false positive and false negative errors, where an item is classified as positive if it is relevant and provides valuable information to an intelligence request, and negative otherwise. The tradeoff between both types of errors, represented visually by the Receiver Operating Characteristic curve, depends on the training and capabilities of the classifiers as well as the classification test performed on each item and the decision rule that separates between positives and negatives. An important question that arises is how to best tune the classification process such that both accuracy of the classification and its timeliness are adequately addressed. An analytic answer is presented via a novel optimization model based on a tandem queue model. This thesis provides decision makers in the intelligence community with measures of effectiveness and decision support tools for enhancing the effectiveness of the classification process in a given intelligence operations scenario. In addition to the analytic study, numerical results are presented to obtain quantitative insights via sensitivity analysis of input parameters.