Application of Machine Learning Techniques on Multivariate Ocean Parameters

MACHINE LEARNING VARIANT ANALYSIS ON OCEAN PARAMETERS

Abstract

Locating potential fishing zones is a requirement for aquaculture. The existence of Potential Fishing Zones is dependent on several ocean parameters. The goal of this paper is to analyze the various techniques to identify the Potential and Non-Potential Fishing Zones based on multivariate parameters like Sea Surface Temperature, Chlorophyll and Salinity. Regression-based model, that is derived from Random Forest methodology has been developed in order to process the dependent parameters, and the outcome is compared with other methodologies namely Support Vector Method (SVM), k-Nearest Neighbor (k-NN), and Decision Trees. The data used for this analysis is the California Cooperative Oceanic Fisheries Investigations (CalCOFI) dataset, which represents the hydrographic data since 1949, of the Californian Current System. The overall efficiency of each method is captured using Accuracy, Prediction Precision, and Area under the ROC Curve (AUC), F1 Score and Recall values. The test accuracy of the proposed system based on Random Forest has been recorded as 96.21 as compared to other methodology. The SVM, k-NN and Decision Tree methods have recorded 79.21, 93.14 and 96.11, respectively. The evidence based on the prediction outcome has affirmed the relationship between chlorophyll and SST, as well as with the Salinity data.