I started data modelling as student back in 2009. I was working on a model where you could feed the contact pattern of a population and it would predict infection spread. I always believed that even a small step in this direction may have a sizeable impact.
After graduation, I realized industry is looking for a different type of data science. The goal there is not to find the fastest way to control an infection, but on how to optimize presenting ads so people buy more. The talent is not spent on using big data to understand why young people in low SES communities pick smoking, but on making political campaign platforms which measure what people love to hear, so that politicians can say the same things on the stage.
To solve this, I started Ethica Data (https://www.ethicadata.com) together with my grad school team. Our intention was to create a behavioural data science platform where the talent and tools in big data can be used to solve problems that matter to society. The focus has been on problems that one way or the other are related to human behaviour.
In this platform you can pick a project and explain why you think it’s important. Then you can ask for data and resources you need for it from your audience, and if they also feel it’s importance, they can help you out.
For example, take behaviour of schizophrenic patients. To understand this, you need data on contact and mobility pattern, and quality of social interaction from a set of schizophrenic and normal population. Then you can train machine learning algorithms to highlight early symptoms. Ethica allows you to borrow data you need from people who meet the criteria and also feel the need, and helps you with the analysis and modelling tools.
We have been working on it for past three years, and there has been some small tractions around it as well. We believe the problem is very important and sooner or later we will see some solution for it, whether from us or others.
In the world of data science, 99% of the problem is defining the problem. There's huge armies of data scientists ready to tackle well defined problems.
How do you hope to do a really good job of defining the problem in a meaningful insightful unconventional way? What are your tactics to do so?
The problem with ResearchKit is that it's hunting for Hawthorne-effect-less data even though it's rife with it.
The issue with a lot of the Validic-style MEMS data is that there's no indication that walking/running/steps counted/ or other key MEMS recorded activities account for a gene regulation pathway?
I haven't seen any papers on meaningful environmental factors being accrued from MEMS collected data.
Neurogenerative disorders can't be modeled easily from early onset symptoms. While from a data perspective it's exciting, many bioinformaticists might disagree. I might be wrong though.
Using a different example could be a good move.
Godspeed.