Statistical pattern recognition in paleontology using SIMCA-MACUP

Abstract

This paper describes a pattern recognition method, Soft Independent Modeling of Class Analogy-Modeling And Classification Using Partial least squares (SIMCA-MACUP), and demonstrates its application to two common paleontological problems—identification and prediction. SIMCA-MACUP performs statistical pattern recognition at four hierarchical levels. At level 1, SIMCA builds disjoint principal component models to characterize multivariate data patterns of several classes. At level 2, new, unknown objects are either classified into one of the classes, or recognized as outliers. At levels 3 and 4, the MACUP method models the relationship between external (dependent) variables and internal (independent) variables, and allows prediction of the values of external variables of new objects which are classified at level 2.The applicability of the SIMCA-MACUP has been demonstrated in a case study of the Pliocene planktic foraminiferal cladeGloboconella.Two morphotypes of the cladeGloborotalia (Globoconella) puncticulataandGloborotalia (Globoconella) inflatawere characterized in a reference sample with SIMCA. Ancestral forms of the two morphotypes were traced through the early phylogenetic history of the clade, resulting in a reconstruction of the branching pattern of the divergence ofG. inflatafromG. puncticulata.The apertural height and apertural width of specimens in the ancestral stocks were then treated as if there were two “external” variables, and predicted using MACUP. The predicted values show slight discrepancy from the observed values. The deviation is size-dependent, indicating a decoupling of the apertural shape from the size of foraminifers during the branching process.