Anomaly Detection Procedures in a Real World Dataset by Using Deep-Learning Approaches

Abstract

Water covers 71% of the Earth’s surface and is vital for all known forms of life. Quality of drinking water is very important. The concentration of major chemical elements under the desirable limit is good for health but an increase in the concentration of the element above the desirable limit may cause adverse effects on human health. Major problems being faced by the world population are due to the presence of excess fluoride, sulfate, chloride, nitrate, and sodium in water. In this paper, we address the problem of changes in the drinking water quality and the crucial task for public water companies to monitor the quality of water. Requirements for drinking water quality monitoring change frequently, e.g., due to contamination by civilization itself or in the supply and distribution network. The proposed methods are K-Nearest Neighbour Algorithm (KNN) and Classification Neural Network based on Logistic Regression for obtaining an appropriate solution in an adequate period of time. Also, the paper compares of the result between the proposed methods and other methods applied in previous work. All experiments are carried out using data gathered from Thuringer Fernwasserversorgung (TFW) water company.