Scholarlink

This paper presents a comprehensive study on classifying depressed and healthy individuals using the Depresjon dataset, which contains motor activity data collected from wearable devices. We prepared six different datasets, including raw data, normalised raw data, PCA-transformed data, and statistical features extracted from the raw data. We trained and evaluated six popular machine learning models and their combinations using a 5-fold cross-validation technique. Our results demonstrate that most models achieved the highest accuracy with the normalised statistical feature dataset. Furthermore, we fine-tuned these algorithms using GridSearchCV and selected the best threshold using the ROC curve. Our findings provide valuable insights into the potential of wearable sensor data for detecting and predicting depressive episodes.

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

Classifying Depressed and Healthy Individuals Using Wearable Sensor Data: A Comparative Analysis of Machine Learning and Deep Learning Approaches

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Faiza Guerrache,
  • David J. Brown,
  • Mufti Mahmud

摘要

This paper presents a comprehensive study on classifying depressed and healthy individuals using the Depresjon dataset, which contains motor activity data collected from wearable devices. We prepared six different datasets, including raw data, normalised raw data, PCA-transformed data, and statistical features extracted from the raw data. We trained and evaluated six popular machine learning models and their combinations using a 5-fold cross-validation technique. Our results demonstrate that most models achieved the highest accuracy with the normalised statistical feature dataset. Furthermore, we fine-tuned these algorithms using GridSearchCV and selected the best threshold using the ROC curve. Our findings provide valuable insights into the potential of wearable sensor data for detecting and predicting depressive episodes.