Innovative Approaches to Gender Classification through Unsupervised Machine Learning and Multi-Activity Fusion

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Gender recognition is crucial for applications in healthcare, sports, rehabilitation, and wearable technology.
• Wearable sensor devices offer a promising avenue for objective human activity analysis.

Purpose of the Study:

• To develop and evaluate a wearable sensor system for gender classification.
• To investigate the effectiveness of unsupervised machine learning for gender recognition based on human activities.

Main Methods:

• Utilized a wearable sensor device with inertial measurement units placed on the upper and lower body.
• Recorded data from seven activities including standing, walking, and climbing.
• Employed unsupervised machine learning algorithms, specifically K-means Clustering and Gaussian Mixture Models (GMM).

Main Results:

• Single-activity gender classification using K-means Clustering and GMM achieved a maximum accuracy of 86.42%.
• Multi-activity gender classification demonstrated superior performance, reaching up to 90.14% accuracy.
• Optimal accuracy was observed with a combination of activities: Romberg test (eyes open), Single leg stance (eyes closed), Walking, and Staircase up and down.

Conclusions:

• Unsupervised machine learning, combined with wearable sensor data, can effectively classify gender based on physical activities.
• Multi-activity analysis significantly enhances gender recognition accuracy compared to single-activity approaches.
• Specific sensor placements and identified behavioral patterns are key to improving classification performance.