Optimized Deep Learning Model Using Nature-Inspired Algorithm for Depression Sentiment Analysis

Introduction: Depression is a prevalent mental health disorder that significantly impacts emotional well-being, cognitive function, and daily activities. Early detection is essential for successful intervention, but conventional diagnostic approaches frequently struggle with identifying issues promptly. Objectives: The research objective is an improved and optimized form of Feature Deep Learning (DL) model using nature-insipired algorithm. It improves the performance of analysis and classification of depressed content from the given online Twitter Dataset. Various objective of this research are: (i) To analyse the existing recent text-based depression analysis methods. (ii) To preprocess and optimize feature-based data using the nature inspired algorithm (embedded-PSO) for the depression dataset. To implement enhanced DL model using RNN-LSTM method for the classification of depression sentiments, and evaluate the different performance metrics and compared with the existing methods, such as CNN-LSTM, Random Forest (RF), Naïve Bayes (NB), etc. Methods: This study introduces a Nature-Inspired Feature-based Recurrent Neural Networks Long short-term memory (NIF-RNN-LSTM) model to detect signs of depression through social media text analysis. The analysis leverages the "Depression: Twitter Dataset + Feature Extraction" dataset from Kaggle, containing 20,000 labeled tweets. Results: The NIF-RNN-LSTM achieved an outstanding accuracy of 96.81%, with a minimal error rate of 3.19%. The efficiency of the proposed model is measured by comparing it to traditional models such as Convolutional Neural Networks (CNN) and Logistic Regression (LR) using standard evaluation metrics, the proposed model consistently outperformed in all evaluation metrics, indicating its robustness and reliability for depression detection tasks. Conclusions: The study presented a NIF-RNN-LSTM model for the prediction of depression, evaluated against CNN and LR models using multiple performance metrics. The results demonstrated that the proposed model consistently outperformed the CNN and LR models across all metrics, showcasing its superior capability for depression classification. With a significantly higher accuracy and lower error rate, the proposed model proved to be more reliable and efficient. Its strong performance in both precision and recall, reflected by the highest F1-score, emphasizes its effectiveness in identifying true positives while minimizing false positives. The proposed model serves as a robust framework for detecting depression and could be expanded in future research to incorporate multi-modal data, enabling broader applications in clinical and real-world settings.