CLASSIFICATION FOR BIG DATA DRIVEN MARINE WEATHER FORECASTING USING MACHINE LEARNING TECHNIQUES

Marine weather forecasting has raised sizeable awareness in numerous ocean related domains. With immeasurable portion of data to handle with, big data problem solving unfolds gateways for abundance of predictions. Machine Learning (ML) is an essential algorithm for big data prediction. However with the increasing size and nature of data i.e., Big Data, Marine Weather Forecasting with Big Data with minimum time, error and maximum accuracy is of major concern to be addressed. In this work, a method called, Perceptred-based Feature and Kriging Gradient Boost Classification (PF-KGBC) is introduced with big data with the objective of improving the prediction performance marine weather with high accuracy and less time consumption. The PF-KGBC method is split into two parts. They are feature selection using perceptron classifier model and classification using Kriging Ensemble extreme Gradient Boost for marine weather forecasting. With the assistance of supervised learning algorithm based on perceptron classifier that involves a functional input represented by vector of numbers belongs to particular class. Based on linear predictor function set of weights are integrated with feature vector to make essential marine weather feature vector. After feature selection process, Kriging Ensemble extreme Gradient Boost Classification is performed with the purpose of forecasting marine weather data. Here, the weak learner is combined to form strong classifier. Kriging regression estimates the dependent data variation when any of factors or values in independent data gets changed. With this accurate marine weather prediction with high accuracy and lesser time consumption is ensured. Meanwhile, we analyzed the results of the proposed PF-KGBC method compared with other conventional methods, and the running performance on the Java platform. The results show that the proposed method achieved satisfactory prediction results and improvements were observed in terms of accuracy, time and error rate considerably.