A Lean and Performant Hierarchical Model for Human Activity Recognition Using Body-Mounted Sensors

Area of Science:

• Human Activity Recognition
• Machine Learning
• Wearable Sensor Data Analysis

Background:

• Human activity recognition (HAR) is crucial for health monitoring and smart environments.
• Existing methods often require complex models and extensive feature engineering.
• Wearable sensors like accelerometers and gyroscopes generate rich data for activity classification.

Purpose of the Study:

• To develop a novel, computationally efficient machine learning algorithm for human activity classification.
• To evaluate the algorithm's performance against existing methods on benchmark datasets.
• To demonstrate the effectiveness of low-complexity models with careful data inspection.

Main Methods:

• Proposed a hierarchical system of logistic regression classifiers.
• Utilized a small set of features extracted from filtered accelerometer and gyroscope signals.
• Tested the algorithm on the DaLiAc (Daily Life Activity) and mHealth datasets.

Main Results:

• The proposed algorithm outperformed previous work on both DaLiAc and mHealth datasets.
• Achieved significant improvements in computational costs with no need for feature selection or hyper-parameter tuning.
• Demonstrated robust performance even with data from only two sensors (ankle and wrist).

Conclusions:

• Low-complexity machine learning models can effectively classify advanced human activities.
• Careful upstream data inspection is key to designing efficient and accurate HAR systems.
• The developed algorithm offers a practical and efficient solution for human activity recognition.