The Intelligent Forecasting Model of Time Series
Sonja Pravilović · Automation Control and Intelligent Systems · 2013
Automatic forecasts of univariate time series are largely demanded in business and science. In this paper, we investigate the forecasting task for geo-referenced time series. We take into account the temporal and spatial dimension of time series to get accurate forecasting of future data. We describe two algorithms for forecasting which ARIMA models. The first is designed for seasonal data and based on the decomposition of the time series in seasons (temporal lags). The ARIMA model is jointly optimized on the temporal lags. The second is designed for geo-referenced data and based on the evaluation of a time series in a neighborhood (spatial lags). The ARIMA model is jointly optimized on the spatial lags. Experiments with several time series data investigate the effectiveness of these temporal- and spatial- aware ARIMA models with respect to traditional one.