Project Title
Classification and Comparative Analysis of In-Hospital Suicidal Behaviors of Patients Using Neural Networks And NLP Techniques
Disciplines
Artificial Intelligence and Robotics | Data Science | Psychiatry and Psychology | Psychology | Public Health
Abstract (300 words maximum)
Suicide, an alarming public health, is one of the top 20 problems in the United States that leaves a lasting impact on families and communities. After a two years decline, a total of 47,000 people committed suicide last year. According to the US CDC report, a person expires every 11 minutes due to attempting suicide. Suicidal behavior information in the health records will help to understand the mental health situation of a patient. By identifying the patients who are ideating and are anticipating attempting suicide using the growing technology, physicians can help the patients' lives by keeping close monitoring. As part of the Phase-I of this project, we built a simple LSTM model on the extracted ScAN [1] (Suicide Attempt and Ideation Events Dataset) data, achieving 94.83% accuracy in predicting the Suicide Attempt (SA) and Suicide Ideation (SI) classes. In the next phase(s), we will apply complex and state-of-the-art model architectures, such as Capsule Neural Networks and EXAM-a three-layer architecture model, for classifying the suicide annotated data.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Md Abdullah Al Hafiz Khan
Classification and Comparative Analysis of In-Hospital Suicidal Behaviors of Patients Using Neural Networks And NLP Techniques
Suicide, an alarming public health, is one of the top 20 problems in the United States that leaves a lasting impact on families and communities. After a two years decline, a total of 47,000 people committed suicide last year. According to the US CDC report, a person expires every 11 minutes due to attempting suicide. Suicidal behavior information in the health records will help to understand the mental health situation of a patient. By identifying the patients who are ideating and are anticipating attempting suicide using the growing technology, physicians can help the patients' lives by keeping close monitoring. As part of the Phase-I of this project, we built a simple LSTM model on the extracted ScAN [1] (Suicide Attempt and Ideation Events Dataset) data, achieving 94.83% accuracy in predicting the Suicide Attempt (SA) and Suicide Ideation (SI) classes. In the next phase(s), we will apply complex and state-of-the-art model architectures, such as Capsule Neural Networks and EXAM-a three-layer architecture model, for classifying the suicide annotated data.