A Micro Neural Network for Healthcare Sensor Data Stream Classification in Sustainable and Smart Cities.

Area of Science:

• Artificial Intelligence
• Internet of Things (IoT)
• E-health
• Smart Cities

Background:

• Smart cities leverage data analysis via sensors and algorithms for efficiency.
• E-health applications using AI and IoT can enhance sustainable urban development.
• Current deep learning models for healthcare sensor data are computationally intensive and difficult to deploy on edge devices for real-time analysis.

Purpose of the Study:

• To address the limitations of existing deep learning models for real-time physiological signal classification on resource-constrained devices.
• To propose a novel, micro time series classification model (MicroNN) suitable for deployment on tiny edge computing devices.
• To demonstrate the application of MicroNN in long-term physiological signal monitoring and its contribution to smart city initiatives.

Main Methods:

• Development of a novel micro neural network (MicroNN) designed for efficient time series classification.
• Deployment of MicroNN on edge computing devices for real-time analysis of physiological signals.
• Comprehensive experimental evaluation of MicroNN's classification accuracy and computational complexity against state-of-the-art methods.

Main Results:

• MicroNN demonstrates superior performance compared to existing state-of-the-art methods.
• Achieved high classification accuracies of 98.4% on the MIT-BIH-AR dataset and 98.1% on the INCART dataset.
• MicroNN exhibits significantly reduced model size and computational requirements, enabling edge deployment.

Conclusions:

• MicroNN is an effective and efficient model for real-time physiological signal classification on edge devices.
• The proposed model facilitates long-term health monitoring, reducing medical and travel expenses.
• MicroNN contributes to the advancement of sustainable and smart cities through enhanced e-health capabilities.