Trained on 200+ years of shark attack and marine weather data, SafeWaters.ai create 7 day attack risk forecasts at any beach. Assess your risk before entering the water. This is great for surfers, local authorities and beachgoers.
Congrats on the launch, very cool! It seems this is only going to become more useful over time and it feels like sharks have definitely become more prominent in the zeitgeist of our day. I'll give it a try. Since shark attacks are so uncommon, how do you overcome the relative low volume of data available to train the models? Also curious to hear a bit more about the big variables that are drivers - water temp, time of year, weather / beach crowdedness?
Its really amazing & heartwarming to see people build products that have a social angle to it
Really appreciate @evanvalenti for picking this up for a greater cause
I really wish this goes a long way in helping humans & sharks, Good luck
While learning machine learning and doing projects like forecasting crime in certain areas, I decided to take the day off and surf. After my surf sesh, I got home to see on the news that while I was out there, there were 8 great whites tagged at the very beach I was surfing at. Then, SafeWaters.ai was born. After Development, our model showed a staggering 83% accuracy in classifying previous attack days/location in our test data as "high risk".
SafeWaters isn't just about saving human lives, but the sharks as well. Current attack mitigation techniques like shark nets kill millions of sharks a year. We're seeking to become the most trusted mitigation method, so unnecessary nets can be removed.
We also pledge to donate 5% of profits to ocean cleanup efforts.
@evanvalenti If you trained your data on previous attack days/locations, then had it assess those days, I would expect it should be pretty accurate for those days. What's the accuracy of predicting a zero attack day as low risk?
@rossdcurrie almost 100 percent, being most days are low risk. There are some low risk days classified as high risk here and there but that's just the nature of the tech.
Like the idea? Maybe you can transition this into an application for surfers in general, providing maps and other values? Like a Nomadlist for surfers!
@lucas_pilzen A colab with the shark tracker app would be sick
Report
Sure, that product is impressive. However, what is the size of your market? I believe your app can be used by a genuine audience. Is my understanding correct?
@kinzarra Yes. Target market is the whole surf industry and regular beach goers who live near/go to beaches with a history of attacks. And we have 0 competition.
Report
Congrats on the launch, very cool! It seems this is only going to become more useful over time and it feels like sharks have definitely become more prominent in the zeitgeist of our day. I'll give it a try. Since shark attacks are so uncommon, how do you overcome the relative low volume of data available to train the models? Also curious to hear a bit more about the big variables that are drivers - water temp, time of year, weather / beach crowdedness?
@adam_fry luckily we've been recording attacks since the 1800's so I had a decent amount of data to work from. There's roughly 30 marine weather data points included in the dataset as well. Date, lat/long are factors as well. I plan on working with marine bio's and shark researchers to dig into the different variables and there impacts. There's also a glimpse of hope for a potential shark week episode next year regarding this.
Report
@evanvalenti awesome, I can definitely see the value in your work and product and wish you continued success as you build out it out - saving lives, both marine and human ones!
Hi friend! Congrats on the 🚀. I have a question, if the shark will attack, but your app told that there is a low risk, will you be responsible for it? I think you should write it under the results!
@anna_kasumova you have to acknowledge a disclaimer before using the app which states that unprecedented attacks can and will occur despite the risk forecast and we are not liable. it's simply a tool to help, not perfect.
Given shark attacks are rare, and practically non-existent for people swimming at the beach, do you not think that apps like this just contribute to the mass-hysteria and negative public sentiment around sharks?
Also, just to be clear, the risk factors are ones that you've determined by training your model on weather data? I assume then it doesn't take into account things like "the person attacked was spear-fishing next to a pack of seals, next to a whale carcass"?
@rossdcurrie lat, long, date/time of year and roughly 30 different marine weather variables are are what the models are trained on. I used unprovoked and provoked attacks in the data as you can't be too sure if the conditions still played a role a provoked attack too. I think some conditions just make them more aggressive in general. Most of the attacks in the dataset are unprovoked so It wouldn't make a big difference in the results, but I will definitely be testing the difference.
Its really amazing & heartwarming to see people build products that have a social angle to it
Really appreciate @evanvalenti for picking this up for a greater cause
I really wish this goes a long way in helping humans & sharks, Good luck
SafeWaters.ai