Scholarlink

This work presents a comparative analysis of the performance of different classic machine learning models for detecting depression from voice signals, aiming to provide a robust methodology for evaluating these approaches. The AVEC2017 dataset was used, considering only the patients’ voice signal data. The models used were Decision Tree (DT), Support Vector Machine (SVM) and Gaussian Naive Bayes (GNB). Also, a study varying the extracted features (namely Insterspeech 2009, IS09, and Interspeech 2010, IS10), the feature selection method (ANOVA and mRMR), and the number of selected features (50, 100, 150 and all features) was carried out. The best results were obtained with SVM, reaching an accuracy of 64.98% and F1-Score of 0.4083 using ANOVA50 + IS10.

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 Comparative Analysis of Classification, Feature Selection, and Feature Extraction Methods for Depression Detection Using Voice Signals

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Matheus Damasceno Prata,
  • Kenji Nose-Filho

摘要

This work presents a comparative analysis of the performance of different classic machine learning models for detecting depression from voice signals, aiming to provide a robust methodology for evaluating these approaches. The AVEC2017 dataset was used, considering only the patients’ voice signal data. The models used were Decision Tree (DT), Support Vector Machine (SVM) and Gaussian Naive Bayes (GNB). Also, a study varying the extracted features (namely Insterspeech 2009, IS09, and Interspeech 2010, IS10), the feature selection method (ANOVA and mRMR), and the number of selected features (50, 100, 150 and all features) was carried out. The best results were obtained with SVM, reaching an accuracy of 64.98% and F1-Score of 0.4083 using ANOVA50 + IS10.