Game Theory based Early Classification of Rivers using Time Series Data

Abstract

An important issue in water quality monitoring of rivers is to identify the river by using the time series data. Water quality involves parameters such as pH, electrical conductivity, dissolved oxygen, turbidity, etc. For real-Time monitoring, such data are collected using sensors from various geographical locations, in different seasons. In this work, our objective to classify the rivers using time series of the parameters, as early as possible. It is difficult to assign the correct label to the river if a time series does not have any information on geographical location. Therefore, we propose an early classification approach to classify the rivers using time series data. We use a probabilistic classifier to obtain the posterior class probabilities to state a tradeoff between accuracy and earliness. A game model is developed to find the optimal number of data points that can provide the desired level of accuracy. We evaluate the proposed approach by classifying the three major rivers of India using time series data. The experimental results illustrate that the proposed approach performs better than the existing approaches in terms of accuracy and earliness. © 2019 IEEE.