A Novel Approach to Predict the Level of Suicidal Ideation on Social Networks using Machine and Ensemble Learning

COVID-19 pandemic has taken millions of lives across the globe. Besides the death toll, the ongoing pandemic has a significant impact on mental health. Researchers contributed in this field revealed that the previous epidemics/ pandemics have a substantial relationship with the elevated rates of suicide. A growing apprehension is that there will be a spike in suicidal cases in the COVID-19 also that will be the next challenge for the world. Social Networking Sites (SNS) are becoming a new way for people to express their thoughts without worrying about the social stigma associated with mental illness. The various risk factors related to suicide like hopelessness, insomnia, anxiety and depression, if anticipated, can help in preventing suicide, thereby avoiding for being the potential victim of this mental disease. This paper focuses on the association between COVID-19 pandemics and suicidal ideation posts/tweets. A novel approach based on machine learning techniques is proposed for detecting and classifying suicidal posts/tweets into different levels of distress. This paper proposes a methodology through which relevant data related to suicidal ideation on social media is extracted, and a hybrid feature engineering mechanism is also proposed for obtaining the useful features from the dataset. The extracted features are then supplied to machine learning and ensemble learning models that automatically classify the risk of suicidal ideation into different levels (Multi-level classification) based upon their severity. Experiments reveal that F-measure is ranging from 0.69-0.98 with the best performance achieved through a Decision tree and Bagging approach. These findings emphasise and encourage the need for automatic classification of suicidal posts in the COVID-19 crisis and the possibility of building and using that machine learning model for immediate suicide risk screening.