Exploring the Effectiveness of Supervised Learning Algorithms for Identifying Suicidal Thoughts in Textual Data
Abstract
Suicide is a major public health concern worldwide. The ability to detect suicidal thoughts in textual data can provide early intervention and support to individuals at risk. In this study, we explore the efficacy of supervised learning algorithms for identifying suicidal thoughts in textual data. We collected a dataset of online forum posts from individuals who had expressed suicidal ideation, and a control dataset of non-suicidal posts. We pre-processed the textual data by cleaning, tokenizing, and vectorizing it is using various techniques. We then trained and evaluated three commonly used supervised learning algorithms: Logistic Regression, Multinomial Naive Bayes, and Linear Support Vector Machines. Our findings show that all three algorithms achieved high accuracy in classifying suicidal and non-suicidal posts, with Logistic Regression and Multinomial Naive Bayes slightly better than Linear SVM. Furthermore, we identified the most significant features for classifying suicidal thoughts, which can inform future research and help develop more accurate models. Our study provides insights into the effectiveness of supervised learning algorithms for identifying suicidal thoughts in textual data and can assist as a source for further research in this important area of mental health.
Keywords
Suicide prevention, Supervised learning algorithms, Natural Language Processing, Feature Extraction