Collective Efficacy of Support Vector Regression With Smoothness Priority in Marine Sensor Data Prediction

Marine data prediction plays an increasingly important role in marine environmental monitoring. The support vector machine (SVM) is viewed as a useful machine learning tool in marine data processing, whereas it is not completely suitable for the abruptly fluctuating, multi-noise, non-stationary, and abnormal data. To address this issue, this paper proposes a novel machine learning framework for marine sensor data prediction, i.e., a support vector regression architecture with smoothness priority. This is a united and consistent system with functions of data acquisition, smoothness, and nonlinear approximation. Here, the smoothness is used to process the outliers and noises of the acquired marine sensor data. Whereafter, a nonlinear approximator based on the SVM is constructed for marine time series prediction. This architecture is the first attempt to consider the collective efficacy of smoother and SVM in marine data processing tasks. The experimental results show that our model significantly surpasses the single SVM in the real-world marine data prediction. Besides, standard statistical evaluation methods, such as QQPlot, PDF, CDF, and BoxPlot, are utilized to verify its superior nonlinear approximation capacity.