Objective: This study aimed to develop and evaluate an artificial intelligence model based on voice features to support the identification of mania, and to build an integrated system for data collection, analysis, and decision support in mental health.
Methods: The training dataset consisted of 46 international patients labeled using the Young Mania Rating Scale, combined with a validation dataset of 4 patient samples from Vietnam. The model was trained using deep learning methods on acoustic features and evaluated using accuracy, ROC, and AUC metrics.
Results: The results demonstrated stable performance on both training and validation datasets, with clear discrimination between different states. When applied to Vietnamese data, the model correctly predicted 2 out of 4 cases, indicating limitations in adapting to regional phonetic variations. An integrated software system was developed, enabling real-time data collection, analysis, and result visualization.
Conclusion: Initial findings suggest the potential application of artificial intelligence in supporting the identification of manic symptoms; however, further data expansion and model optimization are required to improve real-world applicability.