User-Centric Recommendation-Based Approximate Information Retrieval from Marine Sensor Data

Abstract

Due to the wide use of low-cost sensors in environmental monitoring, there is an increasing concern on the stability of marine sensor network (MSN) and reliability of data collected. With the dramatic growth of data collected with high sampling frequency from MSN, the query answering for environment phenomenon at a specific time is inevitably compromised. This study proposes a simple approximate query answering system to improve query answering service, which is motivated by sea water temperature data collected in Tasmania Marine Analysis and Network (TasMAN). The paper first analyses the problems of special interest in missing readings in time series of sea water temperature. Some current practices on approximate query answering and forecasting are reviewed, and after that some methods of gap filling and forecasting (e.g. Linear Regression (LR), Quadratic Polynomial Regression (QPR), Moving Average (MA) and Autoregressive Integrated Moving Average (ARIMA)) are introduced in designing the simple approximate query answering system. It is followed by experiments on gap filling of time series with artificial noise made in the original time series. Finally, the comparison of different algorithms in terms of accuracy, computation time, extensibility (i.e. scalability) is presented with recommendations. The significance of this research lies in the evaluation of different simple methods in forecasting and gap filling in real time series, which may contribute to studies in time series analysis and knowledge discovery, especially in marine science domain.