HYBRID FEATURE BASED PREDICTION OF SUICIDE RELATED ACTIVITY ON TWITTER

Abstract

Globally, the number of suicide deaths rises year, which is a troubling public medical concern. Through this 
experiment, casual, inactive individuals were organically disengaged from online life on Twitter and from 
sharing self-destructive ideas. Initial emotional evaluation of the idle locations was followed by a thorough 
comparison with random factors suggested by space experts. With the increasing normalcy of long-distance 
interpersonal communication platforms, clients have come to rely on these spaces for very sensitive 
conversations, including thoughts of suicide. The tweets are important for research because of the high 
frequency of information appearing in them and the considerable capacity and time constraints that 
computations using them must meet. Emoticons and synonyms features can now be distinguished, and the 
ngram model—a combination of Unigram, Bigram, and Trigram with half breed word reference—is used to 
calculate scores. Using machine learning methods, this model uses the informal points to predict how 
sincere the postings will be. In this study, we also contrast several methodologies such as SVM, NB, and 
RF.