Utilização de técnicas multivariadas na classificação de fases de crescimento da leucena

Abstract

The aim of this work was to identify the growth phases of leucena (Leucaena leucocephala (Lam.) of Wit.), as a function of time, by using of the multivariate analysis techniques. The data set was obtained through an experiment performed at the Experimental Station of Caruaru-PE of the Institute of Agricultural Research - IPA. Besides 20 actual measures of height along time, additional 17 values were used, obtained by interpolation of best fit curves to the Weibull Model, as well as by linear type interpolation, for two distinct treatments groups: according to presence or absence of organic compound residue. The factor analysis was used to reduce the dimensionality of measured data to three factors with eigenvalues higher than unity, explaining 94.60% and 94.30% of the total variation for the treatments with and without organic compound, respectively, using the varimax rotation. The resulting factor scores were subjected to k-means cluster analysis, using previously selected, number of groups k from 3 to 10. The discriminant analysis was then employed to verify the efficiency of clustering of the best classification groups, which was found to be 95.2% and 95.1% of correct classification for the two treatments. The study it showed that the leucena trees which has received with treatment with organic on average attained greater height. The graphical analysis allowed the comparison among the data treatment with and without organic compound treatment, according to groups, in growth phase.