Use of different approaches to model catch per unit effort (CPUE) abundance of fish

Abstract

Fish catch rates are expressed as Catch Per Unit Effort (CPUE) which is a performance index representing the success of fishing from commercial fishery statistics. A three-way comparison of prediction accuracy involving Logistic Regression(LR), Multi-Layer Perceptron (MLP) Neural Networks(NNs) and Classification And Regression Tree (CART) models was performed using a binary dependent variable (CPUE abundance as low or high) and a set of continuous and categorical predictor variables describing seasons, latitude, longitude, gear type, fishing hours and chlorophyll-a concentration. A dataset on CPUE abundance of the Gujarat coastal region during December 2007 to December 2009 was obtained. Overall accuracy (Correct Classification Rate) from NNs and CART models on training were 0.75 and 0.75, respectively and on test data they were 0.73 and 0.67, respectively while by LR they were 0.68 and 0.56 on training and test data, respectively. Present study infers that neither NNs nor CART model showed clear advantage of one over the other. This case study supports the need to test CPUE abundance models with independent data, and to use a range of criteria in assessing model performance. However, the preliminary CPUE Prediction requires multi or related variables in spatio-temporal mode for better CPUE predictions.