The topic relating to abnormal weather captures my attention as I
believe that steps can be taken to curtain the damage that is caused
due to such events. The project is of paramount importance as storms
have become extremely frequent as well as severe in the United States
of America. If the project would be successful, it would be highly
beneficial for thousands of U.S. citizens who are vulnerable to deadly
storms. The goal is to predict the likelihood that any given storm will
produce casualties, as well as identify the five primary features that
may cause a storm to be deadly. The analysis involves creating five
different predictive models and tuning those models until at least one
of them has a classification accuracy of 80% or higher as well as a
sensitivity of at least 70%. By achieving an accurate and reliable
model, our organization will be able to identify characteristics of
deadly storms at a much faster rate. This can help to increase the
warning times issued for any storms that occur in the future. Firstly, a
geospatial map of the U.S. is used to determine which states in the
country have the highest risk of being hit by a dangerous storm.
Secondly, a time series graph is used to evaluate the time(s) of the year
when life-threatening storms are most common. These two methods
are useful for finding the locations that are most vulnerable to deadly
storms, as well as the months during which these storms are most
likely to appear. The final component consists of a text mining study
(word cloud) on the storm event descriptions to determine which
words are frequently used to describe dangerous storms. This will
allow us to identify additional warning signs for deadly storms in the
future.
