Proposed a New Hybrid LOF-ANN Method to Extreme Wave Height Prediction based on Meteorological Data

Extreme wave height prediction is very challenging due to its very high non-stationarity and non-linearity nature. The main aim of the present study is to propose a new hybrid method based on Local Outlier Factor and Artificial Neural Networks classifier, called LOF-ANN, to accurate prediction of extreme wave height occurrence using historical meteorological data. In this study to create models two major hurricanes Dean 2007 and Irene 2011at two locations (NDBC wave buoys stations: http://www.ndbc.noaa.gov) namely; 41004, 41041 in the Gulf of Mexico, is used. TO detect extreme waves, LOF method is used. The outputs of this method are considered as ANN targets. Extreme and normal waves are considered as Class 0 and class 1, respectively. The inputs of ANN models are historical metrological data, including: Wind direction (WDIR), Wind speed (WSPD), Sea level pressure (PRES), Air temperature (ATMP), and Sea surface temperature (WTMP). To create and evaluation of models, the input data sets are randomly divided into training (80%) and test set (20%). The performance of created models is evaluated using three popular criteria Root Mean Square Error (RMSE) and Receiver Operating Characteristic (ROC) and accuracy parameter. The experiment results show that the proposed method is able to predict the occurrence of extreme wave heights with height accuracy (up to 99%).