<![CDATA[Mental health disorders are a growing global concern, with many individuals lacking early detection and appropriate treatment. Mental illness can impact a person’s quality of life and often goes undetected until symptoms worsen. One contributing factor to this problem is the limited ways to detect mental disorders in their early stages. Social media, especially platform X, offers the potential to analyze users’ emotional expressions that may indicate a mental disorder, such as depression or anxiety. Psychological symptoms can be explored more broadly using Natural Language Processing. This study optimizes several text preprocessing techniques to extract meaningful information from social media text. To convert words into numerical vectors, several word embedding methods are used, such as Word2Vec, FastText, and GloVe. Meanwhile, the classification process is carried out using LSTM and Bi-LSTM because they are considered capable of studying data sequence patterns, such as sentence structure, effectively. The results show that the addition of expanding contractions, emoticon handling, negation handling, repeated character handling, and spelling correction in the preprocessing text can improve the model performance. In addition, Bi-LSTM with pre-trained FastText shows better results than the other methods in all experiments, achieving 86% accuracy, 87.5% precision, 84% recall, and 85.71% F1-Score.]]>
