Machine Learning-Based Classification of Suicidal Ideation in Tweets: A Sentiment Analysis Approach

Suicide appears to have reached the highest public health importance, with social media like Twitter providing an infinite source of information that may help in the detection of ideation. This paper describes the development of an effective machine learning-based model for the classification of suicidal ideation in tweets using sentiment analysis. An assembly was done with a dataset of 9,120 tweets; they were labelled according to the suicidal thoughts expressed in them. We presented the assembly and preprocessing of a dataset consisting of 9,120 tweets designated based on indicative or non-indicative suicidal thoughts. The preprocessing steps consisted of normalization, tokenization, stop word removal, and lemmatization. Five ML models, namely Support Vector Machine (SVM), Multinomial Naive Bayes (MNB), Random Forest (RF), Stochastic Gradient Descent (SGD), and Logistic Regression (LR), were further trained on the selected dataset. The models are tested for standard classification metrics: accuracy, precision, recall, F1-Score, and AUC. The best performance for the model was Random Forest with 95.135% and an F1-Score of 95.094%. In conclusion, our study established that an amalgamation of sentiment analysis in ML models can have the potential to detect early suicidal tendencies and that it lays a platform for timely interventions. Future research should be focused on making the dataset even larger and enhancing the detection of sarcasm and jokes in the models.