Performance analysis and comparison of Machine Learning and LoRa-based Healthcare model

Area of Science:

• Biomedical Engineering
• Computer Science
• Data Science

Background:

• Diabetes Mellitus (DM) poses a significant global health challenge, necessitating advanced monitoring solutions.
• The integration of Internet of Things (IoT) and Machine Learning (ML) offers a promising avenue for real-time health monitoring and disease prediction.
• Sustainable development goals emphasize the importance of effective health monitoring systems.

Purpose of the Study:

• To evaluate the performance of a real-time patient data collection model using the Hybrid Enhanced Adaptive Data Rate (HEADR) algorithm within the Long-Range (LoRa) protocol for IoT.
• To assess the effectiveness of Machine Learning classifiers in predicting diabetes severity based on data acquired via the LoRa (HEADR) protocol.
• To compare the performance of various ML classifiers and identify the most effective models for diabetes prediction.

Main Methods:

• Utilized the Contiki Cooja simulator to measure the performance of the LoRa protocol with the HEADR algorithm, focusing on data dissemination and transmission range.
• Implemented diverse Machine Learning classification methods (Random Forest, Decision Tree, k-Nearest Neighbors, Logistic Regression, Gaussian Naive Bayes) for diabetes severity detection.
• Employed k-fold cross-validation to enhance the accuracy of selected ML classifiers.

Main Results:

• The LoRa protocol with the HEADR algorithm demonstrated effective real-time data collection and dynamic transmission range allocation.
• Random Forest and Decision Tree classifiers exhibited superior performance in precision, recall, F-measure, and ROC compared to other models.
• k-fold cross-validation significantly improved the accuracy of k-Nearest Neighbors, Logistic Regression, and Gaussian Naive Bayes classifiers.

Conclusions:

• The proposed IoT system leveraging the LoRa (HEADR) protocol is effective for real-time patient data collection in diabetes monitoring.
• Machine Learning, particularly Random Forest and Decision Tree algorithms, provides accurate prediction of diabetes severity.
• The combination of advanced IoT protocols and robust ML techniques offers a powerful tool for managing and predicting Diabetes Mellitus.