作者: Diana Domańska , Szymon Łukasik
DOI: 10.1016/J.ECOINF.2016.04.007
关键词:
摘要: In the paper methods aimed at handling high-dimensional weather forecasts data used to predict concentrations of PM10, PM2.5, SO2, NO, CO and O3 are being proposed. The procedure employed pollution normally requires historical samples for a large number points in time — particularly forecast data, actual data. Likewise, it typically involves using numerous features related atmospheric conditions. Consequently analysis such datasets generate accurate becomes very cumbersome task. examines variety unsupervised dimensionality reduction obtaining compact yet informative set features. As an alternative, approach fractional distances tasks is considered as well. Both strategies were evaluated on real-world obtained from Institute Meteorology Water Management Katowice (Poland), with extended Air Pollution Forecast Model (e-APFM) underlying prediction tool. It was found that employing distance dissimilarity measure ensures best accuracy forecasting. Satisfactory results can be also Isomap, Landmark Isomap Factor Analysis techniques. These formulate universal mapping, ready-to-use gathered different geographical areas.