A model trained on 6M+ Danish people’s information predicted death with a 78% accuracy.
A new study in Denmark published in Nature Computational Science used machine learning to process the data of life events related to health, education, occupation, income, address, and working hours. The trained model predicts outcomes such as early mortality and personality nuances. Scientists behind the research claim that their algorithm beats others in probing life outcomes by a wide margin.
Link to the Nature study: Using sequences of life-events to predict human lives.
Most notably, the model backed by a decade’s worth of day-to-day records of over 6 million people living in Denmark, predicted deaths with a 78% accuracy. This was tested by prompting the model to provide information on 100,000 individuals from 2016, half of whom have died since then (the model didn’t know that part), and it was mostly accurate in its prediction.
What’s more, this 100,000-person test was conducted on people aged 30 to 55, which is an age bracket where predicting life and death is comparatively harder. Needless to say, this model called “life2vec” can change the game in the insurance industry.
On the personality prediction tests, it was able to comment on a person’s unique traits such as sociability and self-esteem.
The model is not being used anywhere as of now. Google is also working on a similar algorithm.
Featured image: Bikers in Copenhagen, Denmark (Unsplash)
