A Suicidal Ideation Detection Framework on Social Media Using Machine Learning and Genetic Algorithms

Suicide is a serious problem that affects modern societies all over the world, and its prevention is critical. Suicidal ideas are influenced by a variety of risk factors, such as depression, anxiety, hopelessness, and social isolation. Early detection of these risk factors can significantly reduce or prevent suicide attempts. Online platforms, particularly social media, have become a popular outlet for young people to express suicidal ideation. However, detecting and responding to such ideation effectively poses significant challenges in natural language processing (NLP) and psychology. To improve the efficiency of detecting suicidal ideation, this study proposes a novel framework that overcomes prior research limitations, such as feature redundancy and limited relevance to the target class. The framework addresses these issues by extracting meaningful and context-related features from posts that capture contextual and semantic aspects. Furthermore, a genetic algorithm is used to select the most important and relevant features that are strongly associated with the target class while excluding those that are redundant or irrelevant. To evaluate the effectiveness of the proposed framework, the following machine learning classifiers were used to determine whether a post indicates suicidal ideation or not: Random Forest (RF), Naive Bayes (NB), gradient boost classification tree (GBDT), and XGBoost. The proposed framework’s results outperformed previous research, demonstrating the framework’s high efficiency. The best performance in our study was achieved using the Random Forest (RF) classifier, which was applied to the features selected by the Genetic Algorithm (GA) from the linguistic feature set. When evaluated using 5-fold cross-validation, this approach yielded an impressive accuracy rate of 98.92% and an F1-score of 98.92%. These outstanding performance metrics demonstrate the framework’s effectiveness in accurately detecting suicidal ideation.