If you share one single thing from our team in 2026—on social media or per email with your colleagues—please let it be this machine learning competition. It has taken more than half a decade of work to get here, especially for the PhD students Ricarda Proppert and Carlotta Rieble as well as postdoc Rayyan Tutunji, and would not have been possible without an army of volunteers, interns, bachelor and master students, project coordinators, visiting PhD students, and others.
We would love for as many people as possible to participate.
After nearly 5 years of work, our team now finished data collection for my ERC-funded WARN-D project on building a personalized early warning system for depression.
But just because we collected the data does not mean we can fit the best models! So in the spirit of open scholarship, we invite the whole community to participate finding the best predictive model with the highest clinical impact. To do so, we are hosting a machine learning competition, led by my postdoc Rayyan Tutunji, on predicting depression onset. We set up the competition on Codabench, where you can find data for around 1,750 young adults, followed for 2 years. The data include Stage 1, a very comprehensive battery of risk and resilience factors; Stage 2, 3 months of intensive longitudinal monitoring via smartwatches and smartphones; and Stage 3, 8 follow-up measurement points every 3 months, with risk and resilience factors, as well as the main outcome to forecast: depression onset.
Specifically, we share 70% of the data (predictors and onset), as well as the predictors in the 30% holdout. Your job is to provide predictions of depression onset for each person at each measurement point. The deadline is February 27th. The top 3 winning teams will be included as authors on the paper. Details, protocol paper, terms and conditions, codebooks etc. are all available on Codabench. Please don’t hesitate to reach out to me if you have questions.
If this actually works, its a pretty smart way to get free labor! Hahah. If not many sign up I’d suggest a cash prize as additional incentive if there is any sort of budget for that. For what its worth, I could not find the training data file download at any of the links you provided (I did check out the zip file download with the variable names and survey details but it did not include the training data). However this may just be because I did not register a team. I thought it might be fun as a retired computer scientist to play around with the data and attempt to come up with a predictive model but I am not any sort of expert in machine learning and I’m not affiliated with a research lab. I am however interested in machine learning and leveraging AI to help with the software development and training. Gemini came up with some very useful pointers by the way: https://gemini.google.com/share/3da69f7f0243
The modeling approaches look sensible. Cash is tricky given the financial situation of Dutch research institutes after the massive cuts last year, unfortunately. Data are indeed available after you register only, given their highly sensitive nature.
What an excellent idea Eiko! I am in the process of pinching it for a predictive modelling project we are currently conducting :-)
Sweet! Hope it works out ;)