Home // International Journal On Advances in Networks and Services, volume 10, numbers 3 and 4, 2017 // View article
Authors:
Gaetano Zazzaro
Gianpaolo Romano
Paola Mercogliano
Keywords: Data Mining; Forecast Fog; Bayesian Networks, Artificial Neural Networks; Inductive Decision Trees; Model Portability; CRISP-DM
Abstract:
Fog represents high impact atmospherical phenomena especially for aviation. Low visibility conditions severely affect air traffic operations especially during the landing and take-off phases and thereby reducing the capacity of an airport. In particular, in 2001 the Linate Airport in Milan was hit by a disaster, the deadliest air disaster to ever occur in Italian aviation history, due to un-forecasted thick fog. For this reason, improvement of fog monitoring and forecast tool is a challenge topic for the aviation community. Moreover, forecasting fog is an important issue for air traffic safety because adverse visibility conditions represent one of the major causes of traffic delay and of the economic loss associated with such phenomena. In such context, the present work illustrates a Data Mining application for the fog forecasting on a short time range (1 hour) on Linate airport. Indeed two predictive models have been trained using an historical dataset of 18 years of fog observations including many meteorological parameters collected in the Synop message. These models have been made up by applying BayesNet and Neural Network algorithms. The performances evaluation highlights that the complete model shows 90% of instances correctly predicted. Moreover, in order to discover whether predictive models trained on Milan can also be used for forecasting fog events on other geographic sites, a new method to characterize fog events and compare different airport areas is described. Thus, a novel metric is defined, aimed at comparing different sites. This metric is based on the Euclidean distance between performance vectors that are also here defined. Thanks to this metric, we can determine whether a new set of fog observations is compatible or not with Linate fog observations and whether, formally, the predictive models are portable to the new site. Furthermore, we are able to group geographical locations that can be also many kilometers distance away. This work represents a first design step to define the comparative metric. It has been carried on according to the standard process (CRISP-DM) for Knowledge Discovery in Database Process.
Pages: 160 to 171
Copyright: Copyright (c) to authors, 2017. Used with permission.
Publication date: December 31, 2017
Published in: journal
ISSN: 1942-2644