GO-DEPRESSION: Gannet Optimization Codemix Features Based Coordinate Attention CNN-BiLSTM for Penta Depression Detection

Depression is a significant mental health condition affecting millions of people worldwide. Social media platforms, such as Twitter, Facebook, and Instagram often serve as spaces where individuals express their emotions and share personal experiences. However, accurately identifying and categorizing emotional states in code-mixed textual data remains a major challenge due to informal language usage, transliteration, and inconsistent grammar. To overcome these issues, a novel gannet feature optimized deep learning-based penta depression detection (GO-DEPRESSION) framework has been proposed using code-mixed data. Initially, a custom annotated dataset is created using data gathered from platforms like YouTube and Twitter. The NRC Emotion Lexicon is used to label these data and preprocesses them by tokenizing, transliterating, normalizing, and removing noise. A modified FastText embedding is used to extract features and the gannet optimization algorithm (GOA) is used to optimize the features. To perform the classification of the chosen features, a coordinate attention-based convolutional neural network bidirectional long short term memory (CA-CBiLSTM) is used to distinguish the five chosen emotional states which makes it possible to identify the symptoms of depression with the help of the textual information on social media. The framework is written in Python and tested on manually annotated dataset. The GO-DEPRESSION framework is evaluated using specificity, precision, recall, accuracy and F1-score variables. The accuracy of the proposed model is 98.69% and the accuracy of existing MDHAN, DEPTWEET and TS-LSTM is 90.4%, 89.2%, and 88.1%, respectively.