Scholarlink

Objective <p>This study presents GA-SPARF, a wearable Internet of Things (IoT) system for real-time sleep posture monitoring aimed at supporting Gastroesophageal Reflux Disease (GERD) management. Unlike prior studies that typically classify a limited number of sleep postures (e.g., 4–5 classes), GA-SPARF enables real-time classification across 12 distinct sleep positions with high accuracy and low power consumption, thereby enhancing its clinical applicability and scalability for long-term home-based monitoring.</p> Methods <p>The system integrates a single ADXL345 accelerometer positioned on the abdomen, an ESP8266 microcontroller, and Wi-Fi connectivity for real-time data transmission. Data were collected from 20 participants in a controlled environment across 12 labeled postures. Preprocessing involved noise filtering, missing value imputation, and extraction of statistical and kinematic features. A Random Forest (RF) model was optimized using a Genetic Algorithm (GA) and benchmarked against Decision Tree (DT), Gradient-Boosted Decision Trees (GBDT), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) classifiers using 5-fold cross-validation.</p> Results <p>The GA-optimized RF model achieved an average accuracy and F1-score of 99.79 ± 0.02%, outperforming all baseline models. Inference time was 3 ms per prediction, with minimal power consumption. Clinically significant postures—supine (D), right lateral (R), and left lateral (DL)—were classified with 100% sensitivity, positive predictive value (PPV), and negative predictive value (NPV), underscoring the system’s clinical relevance. Despite minor confusion between adjacent postures, the overall classification performance remained consistently high.</p> Conclusion <p>GA-SPARF demonstrates state-of-the-art performance in sleep posture classification using a compact and low-cost setup. Unlike previous studies, it supports a broader set of postures and enables real-time deployment on constrained hardware. This work offers a practical tool for GERD care and lays the groundwork for future innovations in wearable health monitoring.</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

GA-SPARF: Genetic algorithm-tuned random forest for sleep position monitoring using accelerometer

  • 期刊论文

  • 发表日期：

    2026年3月5日

查看全文
  • Hoang-Dieu Vu,
  • Duc-Nghia Tran,
  • Quang-Tu Pham,
  • Duc-Tan Tran

摘要

Objective

This study presents GA-SPARF, a wearable Internet of Things (IoT) system for real-time sleep posture monitoring aimed at supporting Gastroesophageal Reflux Disease (GERD) management. Unlike prior studies that typically classify a limited number of sleep postures (e.g., 4–5 classes), GA-SPARF enables real-time classification across 12 distinct sleep positions with high accuracy and low power consumption, thereby enhancing its clinical applicability and scalability for long-term home-based monitoring.

Methods

The system integrates a single ADXL345 accelerometer positioned on the abdomen, an ESP8266 microcontroller, and Wi-Fi connectivity for real-time data transmission. Data were collected from 20 participants in a controlled environment across 12 labeled postures. Preprocessing involved noise filtering, missing value imputation, and extraction of statistical and kinematic features. A Random Forest (RF) model was optimized using a Genetic Algorithm (GA) and benchmarked against Decision Tree (DT), Gradient-Boosted Decision Trees (GBDT), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) classifiers using 5-fold cross-validation.

Results

The GA-optimized RF model achieved an average accuracy and F1-score of 99.79 ± 0.02%, outperforming all baseline models. Inference time was 3 ms per prediction, with minimal power consumption. Clinically significant postures—supine (D), right lateral (R), and left lateral (DL)—were classified with 100% sensitivity, positive predictive value (PPV), and negative predictive value (NPV), underscoring the system’s clinical relevance. Despite minor confusion between adjacent postures, the overall classification performance remained consistently high.

Conclusion

GA-SPARF demonstrates state-of-the-art performance in sleep posture classification using a compact and low-cost setup. Unlike previous studies, it supports a broader set of postures and enables real-time deployment on constrained hardware. This work offers a practical tool for GERD care and lays the groundwork for future innovations in wearable health monitoring.