A Univariate and multivariate machine learning approach for prediction of significant wave height

Abstract

Predicting significant wave height is important for many engineering projects on the coastline and it is also important for present and future shipping industry. In the present study machine learning and deep learning approaches are employed to predict the significant wave height. The exploratory data analysis is performed on collected meteorological wave data to study the data type and data is filtered and fed in to the machine learning algorithms. Before the analysis the data is divided in to two parts, five years data for training and one-year data for testing. A univariate and multivariate analysis is performed with support vector machine, regression models and deep learning models Long Short-Term Memory (LSTM) and CNN. The common inputs to the models are wind speed, wind direction, gust speed, pressure, sea surface temperature to predict the significant wave height and these inputs are based on correlations with target feature. All the models are trained on 5 years data and tested for the prediction of one-year data. The performance of each model is studied with different error metrics and based on the error metrics the models are compared. The proposed models show an excellent method for predicting the significant wave height.