Paper ID
2954
Paper Type
full
Description
An increasing number of individuals around the world suffer from mental health issues but are unable to access professional help as resources are lacking on a global level. Information technology and especially machine learning has shown great potential as a basis for automated digital services that can act as a support tool for affected individuals. This study presents a state-of-the-art deep learning model for multiclass classification of mental health conditions based on a novel approach of prior classification of symptoms. We contribute to existing research on IS applications in healthcare by improving upon the performance of similar previously reported models, as well as showing that unstructured text can be used to reliably extract not only a primary but also a secondary condition classification. We show that classifying symptoms of individual conditions first and based on that result extracting conditions leads to better model performance.
Recommended Citation
Davcheva, Elena, "Classifying Mental Health Conditions Via Symptom Identification: A Novel Deep Learning Approach" (2019). ICIS 2019 Proceedings. 23.
https://aisel.aisnet.org/icis2019/is_health/is_health/23
Classifying Mental Health Conditions Via Symptom Identification: A Novel Deep Learning Approach
An increasing number of individuals around the world suffer from mental health issues but are unable to access professional help as resources are lacking on a global level. Information technology and especially machine learning has shown great potential as a basis for automated digital services that can act as a support tool for affected individuals. This study presents a state-of-the-art deep learning model for multiclass classification of mental health conditions based on a novel approach of prior classification of symptoms. We contribute to existing research on IS applications in healthcare by improving upon the performance of similar previously reported models, as well as showing that unstructured text can be used to reliably extract not only a primary but also a secondary condition classification. We show that classifying symptoms of individual conditions first and based on that result extracting conditions leads to better model performance.