When Sharks Attack

It’s an image well-known in popular culture, a calm ocean suddenly beset by a terrifying scene of gnashing teeth and thrashing water. The audience gasps in fear as the stricken swimmer fights for their life against a great white shark.

Given this exposure, people may feel a certain trepidation when approaching the realm of these magnificent creatures. But how much do we actually know about shark attacks from recorded data? Is our deep-set fear defensible, or is it only the product of box-office cinema?

In this investigation, we explore the Global Shark Attack File, a compilation of shark attack data from over 100 countries, containing information such as species, location and victim information. Using this data, we’ll answer three posed questions:

1. What proportion of shark attacks are fatal for the humans involved?
2. Can human fatality be predicted accurately using other data from the dataset?
3. How are shark attacks distributed geographically?

The data we used was accumulated from many different researchers globally, and as such there were many inconsistencies in the data entry format. We performed some initial corrections to improve the quality of the data set.

Human Fatality

We found that overall, only 24% of shark attacks result in a human fatality from a set of over 5000 samples. So even if you find yourself in this undesirable situation, you have a better chance of escaping than you may have thought. This lead to the interesting question: are there specific circumstances under which your chances of survival increase or decrease dramatically, that is in what situations are you most at risk of being involved in a fatal shark attack?

We first split the data by species, and there were some clear standouts amongst our species. We can see that the White, Tiger, Bull and Blue sharks are involved in a much higher percentage of fatal attacks, ranging from 18–27%. Perhaps these sharks have a tendency to be more aggressive than other species.

Another interesting split was that of fatality by activity type. We can see from the graph below that the chances of being involved in a fatal attack are over 80% for those people who fell overboard and were attacked. At the other end of the spectrum are Surfing and Kayaking with less than 5% of these attacks being fatal. Presumably, those who fall overboard are more vulnerable than those who are surfing or kayaking and have these items to protect them and provide floatation.

Predicting Fatality

We attempted to predict fatality by applying a simple model to the data. Our results weren’t particularly favourable, we managed to attain an 81% accuracy, but this was largely due to the fact that most of our data landed in the ‘not fatal’ bucket. In fact, we could have attained a 77% accuracy just by predicting every outcome as not fatal!

However, we can highlight the factors that our model indicated as being the highest contributors to the outcome. In order of highest importance, these were:

1. Type: Sea disaster, Correlation: Strong positive
2. Activity: Surfing, Correlation: Strong negative
3. Activity: Boating, Correlation: Strong negative
4. Species: Sand tiger shark, Correlation: Strong negative
5. Activity: Fell overboard, Correlation: Strong positive

Interestingly it seems that if you’re attacked by the Sand tiger shark, the chances of a fatality are greatly reduced!

Geographical Distribution

The heatmap below shows the number of attacks for each species in each location, with only the top 20 locations and species shown. From this, we can see some clear trends. For example, the White shark is prolific and active across a wide geography, in comparison to some species like the Blacktip reef shark whose operation is more localised.