A SPATIO-TEMPORAL BACKPROPAGATION NEURAL NETWORK APPROACH TOFISHERY STOCK PREDICTION

THE PRIMARY OBJECTIVE OF THIS PROPOSAL IS TO DESIGN AND BUILD A SYSTEM FOR FISHERY AND MARINE MAMMAL STOCK PREDICTION USING NONLINEAR SPATIO-TEMPORAL NEURAL NETWORK THEORY. WHILE THE CENTRAL ENGINE OF THE SYSTEM MAKES USE OFNONLINEAR NEURAL NETWORKS FOR PREDICTION, SEVERAL PERIPHERAL MODULES INCORPORATED IN THE DESIGN ASSIST THE USER IN EMPLOYING THE PREDICTOR FOR VARIOUS APPLICATIONS. IN PARTICULAR, OUR APPROACH INVOLVES A USER-FRIENDLY EXPERT SYSTEM AT THE FRONT END TO INTERACT WITH THE USER ON THE DATA DOMAIN, WHILE A RECONFIGURABLE NONLINEAR NEURAL NETWORKFORMS THE CRUX OF A PROBLEM SOLVING SYSTEM. NEURAL NETWORKSHAVE PROVENLY OUTPERFORMED TRADITIONAL LINEAR AND OTHER STATISTICAL PREDICTORS IN THE PAST FOR VARIOUS APPLICATIONS,INCLUDING ENERGY DEMAND PREDICTION (WERBOS 1988). ONE OF THE INHERENT DIFFICULTIES WITH THEM IS THEIR INABILITY TO "EXPLAIN" TO THE USER THEIR DECISIONS. OUR CHOICE OF EXPERTSYSTEMS TO INTERACT WITH THE USER IS WARRANTED BECAUSE OF THE USER-FRIENDLY NATURE OF NATURAL LANGUAGE TYPE INTERACTIONS THAT USERS ARE COMFORTABLE WITH. HENCE AN EXPERT-NEURAL NETWORK PREDICTOR.