«CCUWITY CLASSIFICATION OF THIS ^AOEfWun Data Bnffd) 19. (continued) Network measurements Adaptive identification 20. (continued) 'cluster analysis' procedure was used to. associate a given event with events having similar discriminant patterns. By training on known earthquakes, eight clusters were required to separate earthquakes into groups with similar discriminant patterns. Only one cluster was needed to classify explosions. These results are adaptive, in that no prior knowledge of explosion discriminant patterns were required to obtain these results. Several operational problems were identified by cluster analysis .<» By correcting one of these problems-magnitude scaling-and oy re-running the data base, the same classification performance was\ obtained with only one earthquake cluster. This showed the importance of identifying and eliminating such operational problems. "Qur final results indicated that all of the explosions could be detected as members of a single cluster. However, 141 of the earthquakes remained unidentified and 4% were falsely identified as explosions. This evaluation neglects two explosions both of whiöh were incorrectly edited. Additional work is needed to improve identification of earthquakes by means of multidiscriminant analysis. UNCLASSIFIED SECURITY CLASSIFICATION OF THIS PAGEfWh«! Dal« Enltrtd)