AbstractThe problem is considered of atmospheric meteorological events’ classification, such as sea breezes, fogs, and high winds, in coastal areas. In situ wind, temperature, humidity, pressure, radiance, and turbulence meteorological measurements are used as predictors. Local atmospheric events of 2013–14 were analyzed and classified manually using data of the measurement campaign in the coastal area of the English Channel in Dunkirk, France. The results of that categorization allowed the training of a few supervised classification algorithms using the data of an ultrasonic anemometer as predictors. The comparison was carried out for the K-nearest-neighbors classifier, support vector machine, and two Bayesian classifiers—quadratic discriminant analysis and Parzen–Rozenblatt window. The analysis showed that the K-nearest-neighbors and quadratic discriminant analysis classifiers reveal the best classification accuracy (up to 80% correctly classified meteorological events). The latter classifier has higher calculation speed and is less sensitive to unbalanced data and the overtraining problem. The most informative atmospheric parameters for events recognition were revealed for each algorithm. The results obtained showed that supervised classification algorithms contribute to automation of processing and analyzing of local meteorological measurements.
Increment of diversity with quadratic discriminant analysis – an efficient tool for sequence pattern recognition in bioinformatics
