Date of Award
Master of Science in Applied Mathematics
Department of Mathematics and Statistics
Christine Schubert-Kabban, PhD.
Classification systems are employed to remotely assess whether an element of interest falls into a target class or non-target class. These systems have uses in fields as far ranging as biostatistics to search engine keyword analysis. The performance of the system is often summarized as a trade-off between the proportions of elements correctly labeled as target plotted against the number of elements incorrectly labeled as target. These are empirical estimates of the true positive and false positive rates. These rates are often plotted to create a receiver operating characteristic (ROC) curve that acts as a visual tool to assess classification system performance. The research contained in this thesis focuses on the label fusion technique and the bias that can occur when using incorrect assumptions regarding the partitioning of the event set. This partitioning may be defined in terms of what will be called within and across label fusion. The major goals of this work are the formulaic development and quantification of performance bias between different types of across and within label fusion and analysis of the effects of individual classification system performance, correlation, and target environment on the magnitude of bias between these two types of label fusion.
DTIC Accession Number
Venzin, Alexander M., "Quantifying Performance Bias in Label Fusion" (2012). Theses and Dissertations. 1025.