RI 6898 Linear Discriminant Analysis Of Multivariate Assay And Other Mineral Data

The objective of this report is to explain a statistical method for the summarization, organization, and classification of multivariate assay and other data from the mineral industry. Multivariate data are those characterized by several measurements at each data point, such as assay data from a mixed metal mine, or blast furnace data. A specific method of linear discriminant analysis, the Mahalanobis d2 procedure, is explained, starting from elementary principles; the d2 procedure enables many measurements at many data points to be studied by investigating the interrelationships among the data. The relation of linear discriminant analysis to other statistical methods for treating multivariate data, in particular the eigenvalue technique of factor analysis, is also explained. Example analyses are made on data from the Frisco mine, San Francisco del Oro, Chihuahua, Mexico. The data comprise assay values for silver, lead, copper, and zinc taken at some 19,000 sample points distributed into 91 groups.