Machine learning has been tested at the Finnish Centre for Pensions with encouraging results. Artificial intelligence (AI) managed to identify four out of five retirees on a disability pension already two years before they retired on a disability pension.
We taught a machine learning algorithm to predict retirement on a disability pension based on socioeconomic, earnings and benefit data. With a 78-per-cent accuracy, the algorithm managed to identify the persons who retired on a disability pension two year later.
“Nearly four out of five is a very encouraging result. It is very likely that the accuracy would be improved by adding more social and health data to the algorithm,” says Jarno Varis, mathematician at the Finnish Centre for Pensions.
The sample consisted of anonymous register data of the Finnish Centre for Pensions for 500,000 persons. By comparing the data for the persons who had retired on a disability pension and those whose ability to work was intact, the algorithm learned to identify the variables that predict retirement on a disability pension.
Factors such as high age, repeated sickness benefits and rehabilitation allowances, as well as a reduced earned income were most successful indicators of a forthcoming disability.
“A low educational level, unemployment and being unmarried also contributed to the likelihood of retirement on a disability pension. The factors identified by AI correspond to the results of previous research,” Varis explains.
Artificial intelligence can extend working lives in the future
The machine learning algorithm was most successful at predicted retirement on a disability pension for people suffering from musculoskeletal diseases and disorders in the respiratory organs. It was also better at predicting retirement on a disability pension rather than getting a cash rehabilitation benefits.
Director Mikko Kautto from the Finnish Centre for Pensions sees high potential in artificial intelligence.
“The extensive registers of the Finnish Centre for Pensions offer a good platform for using artificial intelligence in the information industry. Once it get easier to combine more extensive datasets, AI will offer improved tools to help identify risk groups and plan preventive methods,” Kautto says.
Our project was carried out in cooperation with Ilkka Huopaniemi of Siili Solutions. All information processing was done on the platform of the Finnish Centre for Pensions, without cloud computing.