Analysing storm data using highly interactive tools (Q1297893)

The article contains a report on the analysis of hurricane statistical data by the software packages REGARD, MANET, DataDesk and JMP. The data from the US National Hurricane Center (NHC) from 1945 to 1979 contain such variables as date, windspeed, coast distance, landfall (binary: true if the storm hits the US coastline) and so on. These variables are measured several times for each storm. The aim of the analysis is to construct a predictor for landfall variables by other variables. Classical linear and logistic regression models for landfall prediction were fitted using DataDesk interactive methods, but they are relatively poor (\(R^2<16\)\%). Different classes of hurricanes were determined using REGARD and MANET visual tools (including, e.g., traces on maps). Some of the data variables were improved (e.g., the possibility of landfall at the Mexican part of the coastline is taken into account, which was not registered by US NHC). Then a new logistic regression model was fitted with \(R^2=19\)\%.