Can a machine do as well as a human when it comes to detecting suicidal behavior? Well yes, if we are to believe the results of a study appeared in Suicide and Life-Threatening Behavior. A machine-learning computer program recognized a patient’s suicidal nature with 93% accuracy based on analysis of the patient’s own words.
To reach these conclusions, the researchers recruited 379 patients from the emergency room and several medical centers between 2013 and 2015. Each person completed a standardized behavior scale and a semi-structured interview in the form of a conversation. Researchers asked open-ended questions such as “Do you have hope? or “Do you feel angry?” “. By analyzing verbal and non-verbal behaviors during these assessments, scientists were able to classify patients into three categories. People with suicidal tendencies, people with mental disorders without suicidal tendencies, and those who had neither.
The machine almost as reliable as a human
The machine, for its part, analyzed the words of the volunteers, and succeeded in categorizing the suicidal people at 93%. The program also succeeded in differentiating these people from patients simply suffering from mental disorders. This device could be a decision aid for clinicians and caregivers trying to prevent suicides. Use in schools, youth clubs or juvenile justice centers is envisioned by the study authors. For one of them, Doctor John Pestian, “health equipment makes extensive use of technology, but very little is dedicated to the treatment of mental illnesses, whereas algorithms could help caregivers”.
Also to read
One in ten suicide linked to an infection
30% of transgender youth have tried to kill themselves
Suicides at La Poste: experts sound the alarm
