Caprice 1-2
Improving Classification Performance Through Sequences of Classification Tasks (304864)
*Christine M Schubert Kabban, Air Force Institute of TechnologyKeywords: classification, sequence, believe the positive, believe the negative, ROC
Many classification tasks are decomposed into a sequence of classification tasks. For instance, a field of view may be partitioned into natural and man-made objects. After which, the man-made objects are screened to identify an object of interest. These tasks combine classifiers which operate in isolation, yet in fact, perform as a classification sequence. This work examines this scenario, building the classification task as a sequence of classifications. Theory associated with three sequences will be highlighted: Believe the Positive, Believe the Negative and Believe the Extremes. Performance of these sequences will be compared to systems operating separately and the efficiency of these sequences will be demonstrated through the operational cost of the sequence. Further, sequence augmentation will be examined to demonstrate how the classification task may still be completed when information is missing. An example of sequence performance under simulated, theoretical levels of information is examined, and formulas are presented to compute sequence performance. In conclusion, this work demonstrates utility in how sequences fuse information in order to complete a classification task.