Suicidal ideation is a growing concern in the modern digital era, where individuals frequently express their emotions through social media platforms. Detecting early signs of suicidal intent can play a crucial role in preventing self-harm and facilitating timely intervention. This study focuses on classifying social media posts into two categories—potentially suicidal and non-suicidal—to identify individuals at risk. The dataset, sourced from Twitter, is preprocessed to remove noise, normalize text, and handle linguistic variations. Text features are represented using Term Frequency–Inverse Document Frequency (TF-IDF), and multiple machine learning and deep learning models—XGBoost, LSTM, and BERT—are implemented for comparative analysis. The LSTM and BERT models capture contextual and semantic nuances of language, while XGBoost efficiently handles non-linear relationships in textual patterns. The experimental results demonstrate that transformer-based models like BERT outperform traditional classifiers, achieving higher precision and recall in identifying suicidal tendencies. The proposed framework highlights the potential of AI-driven approaches in mental health surveillance, offering scalable, data-driven insights for suicide prevention and awareness programs.

Keywords: Suicidal ideation, Social media analysis, Machine learning, Deep learning, XGBoost, LSTM, BERT, TF-IDF, Text classification, Mental health detection.