A Data which doesn't look your skin color
Since the end of last year, I've started to develop a new application called ClusterStack( http:// clusterstack.io ), a machine learning-driven platform which aims to identify which country you would be more successful to get your visa approved and migrate.
With the manipulation of data and working with various classifications, one thing I've to believe a product like mine MUST do is try to eliminate variables that can discriminate people based on their color, gender, etc ... do the right thing.
BUT, still, in able to help people to migrate, some personal characteristics are still needed in order to create such a feature because this process not only depends on me but on the rules of the country they're willing to move.
So, my question is: nowadays, what are the basics steps a startup must do to ensure they are trying to reduce bias towards data they are building?
5 product ideas with proven demand
Hello, interesting project.
If you don't mind me asking. How do you go about acquiring data to train the model for such a task?
I'm dabbling with machine learning myself. And I always admire cool projects like yours (:
Hello Maltmax, how are you? It's a very interesting question. I've gone through the government's data, in Europe, the US, Brazil, New Zealand, Australia, and Argentina there are laws of data transparency that they have to release for the public. In Peru, there is no much data provided by the government, such as Asian and Middle East countries. Because of the specifics of my project I have to rely more on government public data and information to train my model. So for you, if you have a very specific problem you need data that's not there, the answer would be to capture your own data with questioning forms, there is researchers you can try to hire like in Fiverr you can gather your data.