Clustering for Classification

"We propose to develop clustering techniques for classification of BMD targets. Model-based classification is vulnerable to unknown threats and variations on known threats. Unsupervised classification approaches have been given little consideration within BMD research. A system using a combination of supervised (model-based) classification and unsupervised (cluster-based) classification would be far more robust. We propose to investigate ways in which these two different approaches can be hybridized to dealintelligently with situations in which a solely model-based approach would fail but a solely clustering approach would lose out on useful prior knowledge. The proposed scheme for hybrization builds on techniques we have recently developed for bias removal during work funded by project Hercules. The ability to make correct classification between RV and non-RV is a key requirement in BMD. The benefit in improving classification is an immediate corresponding improvement in system performance. The primary commerical applications of this researchwould be inclusion in five MDAPs, THAAD, NTW, NAD, PAC-3 and MEADS. There are many wider applications in the DoD and private industry for methods that combine model-based and cluster-based classification."