Fingerprints have two broad and disparate but complementary uses. In biometrics, fingerprint matching enables individual identification. In anthropology, fingerprint characteristics describe and differentiate related groups. This research bridges those uses and provides a unique contribution by showing how to use fingerprint analysis to associate a person to one of two groups. Broadly, the groups are restricted to those in which fingerprints present descriptive attributes based upon a degree of common genetic material generated by consanguinity, endogamy, or other factors engendered by heredity or disease.

In comparative dermatoglyphics, anthropologists have shown that fingerprints from endogamous groups display common dermatoglyphic attributes and that the dermatoglyphic differences between groups are statistically significant. This dissertation uniquely employs the fingerprint attributes found useful in comparative dermatoglyphics to associate an individual to one of two endogamous groups.

The process to generate person-to-group associations comprises a number of steps. First, anthropologic dermatoglyphic studies were reviewed to determine fingerprint attributes valuable in differentiating groups. Next, feature vectors composed of the fingerprint attributes describing each person formed a learning set with two classes, one for each of the two endogamous groups. Multiple classifiers were developed then tested using various feature vector constructs in the learning set coupled with a leave-one-out method to determine classification accuracy.

Classifier performance varied with vector structure. When multiple vectors represented a person and separate vectors described each finger, then the classifiers performed poorly. Performance moderately improved by applying various per-finger vector structures and weighting methods. Classification accuracy significantly improved when representing a person with a single, long vector containing per-person and per-finger pattern and ridge count information.

A single, long vector structure enabled approximately eighty percent classification accuracies in associating a person to one of two groups, and various group compositions bolstered the results. When the two groups comprised persons from two Iraqi cities, then the classifier primarily used ridge count measures to achieve a classification accuracy of 80.8 percent. When the combined Iraqi cities formed one group and a collection of Mayan Indians formed the second, then the classifier used pattern as the primary population separator supplemented with ridge count measures to achieve a classification accuracy of 77.4 percent. These classification accuracies, reinforced by varied group compositions, establish an ability to infer individual to group associations for endogamous populations using dermatoglyphic characteristics.