Modelling Chlorophyll-a Concentration using Deep Neural Networks considering Extreme Data Imbalance and Skewness

Abstract

Algal bloom has been a serious problem, as some of algae such as cyanobacteria produce toxic wastes. Chlorophyll-a has been one of the primary indicator of algal bloom; however, it is difficult to model to forecast due to scarceness of the events. Since canonical machine learning algorithms assume balanced datasets, data imbalance of the Chlorophyll-a concentration must be visited for accurate prediction. In this paper, we present a convolutional neural network model to predict Chlorophyll-a concentration, handling its data imbalance and skewness. The experiment results show that proper data transformation and oversampling can improve prediction accuracy, especially in rare-event regions.