Citation

Kumar ER, Venkatram N. Soft Comput. 2023; ePub(ePub): ePub.

Copyright

(Copyright © 2023, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s00500-023-08095-y

PMID

unavailable

Abstract

Online networking sites are increasingly being used to communicate suicidal thoughts. Suicide research and policy are best served by monitoring such conversations and exchanging information of significant relevance for both individual prevention and population-wide goals. As a result, multiple studies discovered that monitoring social media postings might help identify those who are thinking about suicide. Finding and understanding patterns of suicidal thoughts, however, is a difficult undertaking. As a result, it is vital to construct a machine learning system for automatic early identification of suicide thoughts or any sudden changes in a user's behavior by evaluating his or her postings on social media. This may be done by comparing the user's previous behavior to his or her current conduct. In this study, a Twitter dataset containing features like age, gender, follower, following, drunk abuse, hours, and category is utilized to predict and analyze suicide behavior. The class containing the values has the attribute name category normal, low and high denoting the levels of suicide risk, and attributes comprise categorical and numerical data. The Twitter dataset is categorized using a proposed rule-based algorithm by utilizing the quicksort method to determine the best split point from a chosen attribute. The tree's split points classify data, and a decision tree computes all node path distributions. The records with missing values are calculated as part of the missing count after all the class values have been counted. Based on class values, the accuracy of the user age level prediction analysis is performed from the node distribution. Male users had a 32% high suicide risk compared to female users' 24% for age groups under 37. Female Twitter users who are older than 37 have an average suicide risk level of 25% compared to male users who have a prediction level of 20%. It has been observed that the proposed model, when compared to the existing model, obtained a better accuracy for values ranging from 76 to 90% for 7 distributions and more than 90% for 25 node distributions out of 33 nodes. Twitter's benchmark for suicide detection in online social networks is successful, according to the experimental research. The suggested model yields high average accuracy of 70%, according to experimental findings.

Language: en

Keywords

Decision tree; Sorting technique; Suicidal prevention; Twitter