Activity recognition using hybrid generative/discriminative models on home environments using binary sensors.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Activity recognition in smart environments is crucial for monitoring elderly health status.
• Previous studies have explored various methods for activity recognition, but challenges remain in home settings.

Purpose of the Study:

• To develop and evaluate a hybrid machine learning approach for accurate activity recognition in a home environment.
• To improve the health status monitoring of the elderly through enhanced activity recognition.

Main Methods:

• Utilized Artificial Neural Networks (ANN) and Support Vector Machines (SVM) as discriminative models.
• Integrated ANN and SVM output scores as observation probabilities within a Hidden Markov Model (HMM) framework.
• Evaluated the hybrid ANN-HMM and SVM-HMM models against classical activity recognition methods using five real-world datasets.

Main Results:

• The hybrid ANN-HMM and SVM-HMM models demonstrated significantly superior activity recognition performance compared to traditional methods.
• The improved performance achieved a statistical significance level of p < 0.05.
• The hybrid approach proved more effective for activity recognition in the targeted home domain.

Conclusions:

• Hybrid machine learning models combining ANN/SVM with HMM offer a robust solution for activity recognition in smart home environments.
• This approach significantly enhances the accuracy of recognizing daily living activities for the elderly.
• The findings support the suitability of this hybrid methodology for improving elderly health monitoring systems.