Using Multivariate Statistical Analysis to Validate a Geochemical Typology Model: A Case Study of the New Model Proposed for Paraná Igneous Province in the State of Paraná, Brazil
Alexandre Cancian Bajotto, Anselmo Chaves Neto, Otavio Augusto Boni Licht

Abstract
In 2018 a new rock classification model for Paraná Igneous Province (PIP) was proposed using four discriminant variables: SiO2, Zr, TiO2, and P2O5. This research introduces the Multivariate Analysis as a technique to verify and check this new model. The study used a subset with samples from boreholes in the state of Paraná, Brazil. The Subset is a matrix of 1,030 observations x 73 geochemical variables and Factor Analysis reduced the subset to a Factor Scores matrix of 1,030x24. Then, Cluster Analysis grouped the Factor Scores in 5 clusters using K-Media Method, followed by a Pattern Recognition and Classification Analysis using Neural Networks to corroborate the clustering. The Principal Component Analysis extracted Principal Component Values and Weights from the geochemical types proposed by the 2018 model. Finally, the Canonical Correlation Analysis correlated vector U1 of Factor Scores with V1 of Principal Component Values and Weights and the result was the validation of the new classification model.

Full Text: PDF