Efficacy of swarm-based neural networks in automated depression detection

As depression becomes a global pandemic, this research paper presents a comprehensive study for depression diagnosis using a custom-crafted deep learning model optimized with various swarm intelligence algorithms. Three different optimization algorithms-dragonfly algorithm, Firefly Algorithm, and Moth Flame Optimization Algorithm-have been employed for feature selection and dimensionality reduction. The experiments were done with the DAIC-WOZ corpus, which is a benchmark dataset for depression detection. In the first experiment, the dragonfly algorithm was used and gave a macro F1 score 0.76, recall of 0.80, and precision of 0.74. The second experiment was the implementation of the Firefly Algorithm, which outperformed the first with a macro F1 score of 0.86, recall of 0.88, and precision of 0.85. The third experiment based on the Moth Flame Optimization Algorithm achieved a macro F1 score of 0.80, a recall rate of 0.79, and a precision of 0.72. Next, the models were put through two additional datasets to test how well they generalize. Therefore, the model achieved 0.92 F1 on the CMDC dataset and 0.82 on the MODMA dataset, establishing strong performance across different distributions of data. The results highlight how the integration of deep learning techniques with metaheuristic optimization algorithms can provide optimal and reliable depression diagnosis and thus create promising directions for further research in this area.