Children's Activity Classification for Domestic Risk Scenarios Using Environmental Sound and a Bayesian Network

Area of Science:

• Computer Science
• Pediatrics
• Public Health

Background:

• Domestic accidents are a significant global health concern for children.
• Current children's activity classification methods using wearable sensors are prone to errors.
• Non-invasive data sources are needed for reliable child monitoring systems.

Purpose of the Study:

• To propose and evaluate environmental sound as a data source for children's activity classification.
• To develop models for recognizing activities that may trigger domestic accidents.
• To enhance child monitoring systems with a reliable, non-invasive approach.

Main Methods:

• Utilized environmental sound for feature extraction.
• Implemented Akaike criterion and genetic algorithms for feature selection.
• Generated classification models using Naive Bayes, Semi-Naive Bayes, and Tree-Augmented Naive Bayes classifiers.

Main Results:

• Models combining feature selection and Bayesian network classifiers achieved over 97% accuracy.
• Demonstrated the effectiveness of environmental sound for activity classification.
• Successfully identified potentially hazardous activities related to domestic accidents.

Conclusions:

• Environmental sound is an efficient and reliable data source for children's activity classification.
• The proposed method significantly improves the recognition of accident-triggering activities.
• This approach offers a promising non-invasive solution for child monitoring and accident prevention.