Anomaly Detection for Environmental Data Using Machine Learning Regression

Abstract

Environmental data exhibits as huge amount and complex dependency. Utilizing these data to detect anomaly is beneficial to the disaster prevention. Big data approach using the machine learning method has the advantage not requiring the geophysical and geochemical knowledge to detect anomaly. This paper using the popular support vector regression (SVR ) to model the correlation between factors. From the residual of the regression, it develops a statistical method to quantify the extremity of some abnormal observed data. A case study is proposed to demonstrate the developed methods.