Scholarlink

<p>Detecting depression from speech signals is challenged due to inaccurate predictions, uncertainty, and a lack of interpretability. In this paper, a depression detection model is developed with a squeeze excitation transformer model and optimization approaches. Initially, the input signal is collected from a publicly available dataset, then pre-processed to produce accurate detection results. Then these signals are converted into images using the Markov transition field, the Recurrence plot, and the Gramian angular field. From the image, features are selected using the hybridized genetic golden jackal optimization algorithm, crested porcupine optimization algorithm, improved chimp optimization algorithm, and improved coot optimization algorithm. Depression detection is enabled with a squeeze excitation transformer model, and the features are visualized using Shapley Additive exPlanations. Accuracy, precision, recall, f1-score, specificity, TPR, and FPR values obtained with the proposed approach of HGGJOA optimization for the DAIC-WOZ dataset are 0.9802, 0.9681, 0.9621, 0.9651, 0.9871, 0.0128, and 0.9525, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A meta-heuristic optimization framework with explainable AI and SE-transformer integration

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Uma Jaishankar,
  • Jagannath H Nirmal

摘要

Detecting depression from speech signals is challenged due to inaccurate predictions, uncertainty, and a lack of interpretability. In this paper, a depression detection model is developed with a squeeze excitation transformer model and optimization approaches. Initially, the input signal is collected from a publicly available dataset, then pre-processed to produce accurate detection results. Then these signals are converted into images using the Markov transition field, the Recurrence plot, and the Gramian angular field. From the image, features are selected using the hybridized genetic golden jackal optimization algorithm, crested porcupine optimization algorithm, improved chimp optimization algorithm, and improved coot optimization algorithm. Depression detection is enabled with a squeeze excitation transformer model, and the features are visualized using Shapley Additive exPlanations. Accuracy, precision, recall, f1-score, specificity, TPR, and FPR values obtained with the proposed approach of HGGJOA optimization for the DAIC-WOZ dataset are 0.9802, 0.9681, 0.9621, 0.9651, 0.9871, 0.0128, and 0.9525, respectively.