Human Activity Classification Using Multilayer Perceptron

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Wearable Technology

Background:

• Smart homes are increasingly prevalent, offering convenience and automation.
• Integrating smart home data with advanced processing can yield insights into resident well-being.
• Current smart home analytics often focus on occupancy and fall detection.

Purpose of the Study:

• To advance smart home data analysis beyond traditional monitoring.
• To develop a system for recognizing multiple human activities within a smart home environment.
• To leverage wearable sensor data for enhanced well-being insights.

Main Methods:

• Utilized a multilayer perceptron neural network for human activity recognition.
• Employed data from wrist- and ankle-worn sensors.
• Implemented rigorous cross-validation and scoring evaluation methods.

Main Results:

• Achieved very high recognition accuracy for multiple human activities.
• Cross-validation demonstrated accuracy levels exceeding 98% across all models.
• Scoring evaluations showed only a minor average accuracy reduction of 10%.

Conclusions:

• The developed neural network models are highly effective for human activity recognition in smart homes.
• This approach significantly enhances the potential for monitoring resident well-being through smart home technology.
• The findings support the expansion of smart home data analysis for comprehensive inhabitant monitoring.