Suicidal Ideation Detection Using Machine Learning Based Hybrid Techniques

Abstract

The rapid growth of technology and social media platforms allows everyone to express their opinions and emotions and share them with millions. Online Social Networks (OSN) such as Facebook, Twitter, Instagram and WhatsApp are becoming data-sharing platforms. As the number of people participating in the virtual channel increases, the much-unstructured text is generated. These texts help understand the user's state of mind and predict the level of depression or Suicidal Ideation (SI) of the person. Yet, identifying and understanding patterns of SI can be difficult. Identifying and comprehending the complex risk factors and warning signs that lead to suicide is the most challenging part of suicide prevention. To combat this problem, Machine Learning (ML) based hybrid techniques like Support Vector Machine (SVM), Random Forest (RF) and Prism for suicide ideation prediction on social media posts. Firstly, the preprocessing stage eliminates the unwanted symbols, stemming and tokenization. Then we apply Term Frequency-based Information Gain (TFIG) feature extraction to select essential features of suicidal thoughts. Finally, ML-based Hybrid models with Natural Language Processing (NLP) categorize suicide or non-suicidal tweets. The outcome of the simulation analysis is that the proposed hybrid models achieved higher accuracy by observing people's emotions from their text than other methods.