Adaptive Early Classification of Time Series Using Deep Learning

Abstract

Early Classification of Time Series (ECTS) is a process of predicting the class label of time series at the earliest without observing the complete sequence. Time Series data is a collection of data points over time, and a decision has been made based on a complete sequence. However, early decision based on partial information is beneficial in time-sensitive applications. ECTS is an emerging research area with multiple applications in various domains such as health and disease prediction in medicine, Quality and Process Monitoring in Industry, Drought and Crop monitoring in agriculture. In this paper, we propose an adaptive early classification model composed of two components. The first component is the base classifier, which has been designed as a hybrid model of Convolutional Neural Network and Recurrent Neural Network. The Second component is the decision policy designed for adaptive halting capabilities, which has been defined as a reinforcement learning agent to determine when to stop and make a prediction. We evaluated our model on publicly available different kinds of time-series datasets. The proposed method outperformed the state-of-the-art in terms of both accuracy and earliness.