MBOSS: A Symbolic Representation of Human Activity Recognition Using Mobile Sensors

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human activity recognition (HAR) systems using smartphone sensors are valuable for monitoring behavior.
• Existing HAR systems face challenges with high computational resource demands (memory, processing).
• Current supervised HAR methods rely on manual feature extraction, requiring expert knowledge.

Purpose of the Study:

• To propose a novel, efficient method for human activity recognition (HAR).
• To overcome the computational limitations of existing HAR systems.
• To maintain or improve accuracy compared to conventional HAR techniques.

Main Methods:

• Development of the Multivariate Bag-Of-SFA-Symbols (MBOSS) method.
• Utilizing symbolic representation algorithms for activity recognition.
• Evaluation on three public datasets.

Main Results:

• MBOSS demonstrated superior performance across key metrics.
• Significant improvements in processing time and memory consumption were observed.
• Accuracy levels were comparable to traditional time and frequency domain feature-based systems.

Conclusions:

• MBOSS offers a more efficient approach to HAR.
• The method effectively reduces computational resource requirements.
• MBOSS presents a viable alternative for developing practical HAR systems.