The Accuracy Analysis of Different Machine Learning Classifiers for Detecting Suicidal Ideation and Content

Abstract

Suicide is the matter of purposely causing one’s death and suicidal ideation refers to thoughts or preoccupations with ending one’s own life. Studies have explored verbal and written communications related to suicide, including analyzing suicide notes, online discussions, and social media posts to identify linguistic and content markers that may help in early detection and intervention. The primary purpose of this study is to detect signs of risk of suicide/self-harm in social media users by investigating several frequency-based featuring and prediction-based featuring methods along with different baseline machine learning classifiers. The algorithms applied for analysis are Decision Tree, K-Nearest Neighbors, Random Forest, Multinomial Naïve Bayes, and SVM. Our experimental results showed that the best performance is obtained by the FastText embedding with SVM model having the highest accuracy of 93.76% which outperforms other baselines. The aim of this work is to learn the significance of analysis and do a comparative study of algorithms to find the best suited algorithm.