11 January 2016 - 23:57Allergic Rhinitis (“Why do I always sneeze?!”) research project using Machine Learning
Since April my wife (@fluffyemily) and I have been running a research project around her allergies. She sneezes all year and we’re trying to figure out the cause. Allergic Rhinitis affects 10-30% of Westerners, in Emily’s case it is all-year so it isn’t just pollen related. We figure that a good data-collection process coupled with robust analysis might reveal some of the causes of sneezing such that Emily’s in better control of her Rhinitis.
Emily’s a senior iOS developer with Mozilla, she wrote an open source App for her iPhone to log her sneezes, antihistamine use and interactions with “things” like animals. The App gives us a time-stamp and geolocation. Since she’s mostly in London we’ve got a rich source of events to join to other datasets.
This post is just to put down a marker. I’ve made some progress using Machine Learning to predict when an antihistamine might be used. Currently I can out-predict a Dummy (majority-class) classifier using many cross-validation runs, this is hardly brilliant but we didn’t expect diagnosing a long-term allergy to be a simple affair! Exploratory data analysis on the data shows lots of interesting behaviours, I hope to talk about some of these in the future.
We’ve tried (and so far rejected) air-born particulates as a reason for her allergies via Kings College LondonAir data (thanks!). Weather data is more promising using a local wunderground station (Emily seems to be a little sensitive to humidity and windspeed). I’ve recently started work on MyFitnessPal logged data (the Python 3.4 port was thankfully easy) to start to look at alcohol (a known histamine modifier) and possibly other food.
Behind the scenes I’ve got a collaborative group (thanks Frank and Giles!) in Slack and a private github repo, I plan to talk a little on how this works. I think talking about ways we can collaborate on research projects has value, anything that helps us move on from just working in an office seems like a good idea.
If you’re interested in hearing updates about this project and maybe getting involved to log your own allergy data, join this email announce list. Your email will be kept private, I’ll just send you an email every now and again when we’ve made some progress (which will probably appear here) and when we need volunteers.
Ultimately we’d like to help predict the causes of allergies for other folk. We’ve been talking about this for around 2 years, it is encouraging to see research like this pointing to the use of ML to predict and model the body’s behaviours.
Ian applies Data Science as an AI/Data Scientist for companies in ModelInsight, sign-up for Data Science tutorials in London. Historically Ian ran Mor Consulting. He also founded the image and text annotation API Annotate.io, co-authored SocialTies, programs Python, authored The Screencasting Handbook, lives in London and is a consumer of fine coffees.