A simple method for classifying data in relative growth analysis of crustaceans

Abstract

Abstract Relative growth in crustaceans is a topic of quite high importance that allows, among other things, to estimate their size at the onset of morphometric sexual maturity. In this process, it is usual to start from a transition point and then generate regressions that reflect phases of the sexual development of the individuals (i.e., mature and immature). There are several statistical methods that allow data to be classified into subsets, but according to the specific growth pattern at issue, those methods may or may not be usable. This paper proposes a simple method to classify data into two subsets, viz., when they are overlapping over a wide range of sizes. The actual procedure consists of a linear regression that divides the data. Subsequently, a linear regression is applied to the groups generated by adjusting the parameters through maximum log-likelihood. The observed values will be classified according to the smallest residual difference generated with each regression line. The proposed method was tested by separating the sexes of Goyazana castelnaui, using real data. The efficiency of the method was analysed based on the percentage value between the number of total data and the number of correctly classified data. Additionally, the k-mean cluster was used as a conventional method, the results of which were reclassified by a linear discriminant analysis. The efficiency of the proposed method was >80% while that of the conventional method was >60%. The values misclassified by the proposed method were mixed with those of the opposite sex, so it was expected to fail in those cases. The proposed method is a simple alternative that can serve as a basis for subsequent morphometric analysis, especially for acquiring an initial insight in the structure of a dataset collected for a study of relative growth.