Physical Human Activity Recognition Using Wearable Sensors.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Machine Learning

Background:

• Human activity recognition (HAR) is crucial for health monitoring and assistive technologies.
• Wearable inertial sensors offer a non-invasive method for collecting human motion data.
• Evaluating classification techniques is essential for accurate HAR systems.

Purpose of the Study:

• To compare the performance of various supervised and unsupervised classification techniques for HAR using wearable inertial sensor data.
• To identify the most effective classification algorithms for distinguishing daily living activities.

Main Methods:

• Utilized three inertial sensor units placed on the chest, right thigh, and left ankle of healthy subjects.
• Compared four supervised (k-NN, SVM, GMM, RF) and three unsupervised (k-Means, GMM, HMM) classification techniques.
• Employed feature selection using a Random Forest-based wrapper approach.

Main Results:

• The k-Nearest Neighbor (k-NN) classifier demonstrated superior performance among supervised methods.
• The Hidden Markov Model (HMM) achieved the best results in the unsupervised classification context.
• Performance was evaluated using metrics like correct classification rate, F-measure, recall, precision, and specificity.

Conclusions:

• k-NN is highly effective for supervised HAR with wearable sensors.
• HMM provides the best performance for unsupervised HAR in this experimental setup.
• The study provides insights into optimal classification strategies for wearable sensor-based activity recognition.