<![CDATA[Linguistic digital footprints on social media platforms have transformed into psychological biomarkers that offer real-time mental health screening opportunities. However, the main challenge in automated clinical text processing is semantic ambiguity and class imbalances (imbalanced datasets) that lead to predictive bias in conventional models. This study aims to develop a multiclass classification system for seven mental health categories (including Suicidal, Biopolar, and Anxiety) using a Deep Learning approach. The proposed method integrates Bidirectional Long Short-Term Memory (Bi-LSTM) architecture with Word Embedding to capture the context of a two-way narrative, as well as apply Class Weighting techniques to address data distribution disparities. The experiment was conducted on a public dataset containing 52,680 text entries. The test results showed that the model was able to overcome the majority bias very effectively, as evidenced by the high Recall values in critical classes such as Biopolar (0.84) and Suicidal (0.76), as well as an F1-Score of 0.91 in the Normal class. This study concludes that the Bi-LSTM architecture with adaptive class weighting is able to be an early warning system that is sensitive to high-risk disturbance signals.]]>
Copyrights © 2026
