Inspiration
Research findings suggest that mobile phone data, such as location, usage, communication patterns, etc. can predict the likelihood of the outbreak of a depression. Depression is a major illness in modern society causing massive social and individual problems. Early detection and intervention can improve the wellbeing of affected individuals dramatically.
Papers:
- http://www.jmir.org/article/viewFile/jmir_v17i7e175/2
- http://repository.cmu.edu/cgi/viewcontent.cgi?article=1272&context=hcii
What it does
Our system learns parameters from a beginning data set of mobile data that. From there it will predict a probability score for a person's depression depending on their current data (for example of the last week/ last day). If the probability score exeeds a threshold value, e.g. 0.7, the user will get a automated phone call of our help hotline.
This phone call will evaluate the status of the person, by conducting a short depression test. If the test is positive, i.e. the individual shows strong symptoms of a depression, the user will be forwarded to a professional therapist. If the test is negative nothing happens. In both cases the new labeled data point will be added to the training set of our model. Thus the model will improve its accuracy, the more users there are.
How we built it
We separated tasks:
- built the core-machine learning model
- generate a random dataset, with a beginning set and a stream of new data points
- embedding the model into twilio
- build a dashboard for presentation
Challenges we ran into
Our idea is based on scientific correlations, that we could not proof, because of lack of a dataset. We therefore had to create our own random dataset in order to design our model. This was a bit tricky, as we had to model it as close to the dataset of the studies as possible.
What we learned
A lot about the usage of scikit and twilio
What's next for Depression Predictor
We will market it and make tons of $$$$$ :-P
Log in or sign up for Devpost to join the conversation.