Statistical Analysis Based on Lake Michigan Fish Acoustic Data Using LASSO Method

Abstract

LASSO method is one of the most popular and more extensive regressions. It has been applied to many fields. However, it is rare seen to research with complicated big data in biology. This paper is to apply LASSO method to Lake Michigan Fish acoustic data. The main techniques include: Elastic Net selection, which tests validation from the average square error (ASE) to predict the error for the model by computing separately for each of these subsets; defaulting group LASSO to test multiple parameters by splitting a couple constituent parameters, such as successive intervals, multiple continuous depth layers, to estimate the Schwarz Bayesian information criterion (SBC) to find the lowest value for the model; The adaptive LASSO selection, which is applied to each of the parameters in constructing the LASSO constraint for weights, that is, the response y has mean zero and the regressor x are scaled to have mean zero and common standard deviation. The empirical results show that the fish density (Y) has strong relationships with area backscattering coefficient (PRC_ABC), secondly, significant interactions with PRC_ABC and Exclude below line depth mean), among PRC_ABC, fish density in the intervals and layers of acoustic survey transect of Lake Michigan.