myBlackBox: Blackbox Mobile Cloud Systems for Personalized Unusual Event Detection

Area of Science:

• Mobile computing
• Data fusion
• Machine learning

Background:

• Personal event detection is crucial for user-centric mobile applications.
• Existing systems lack efficient fusion of multimodal sensor data.
• Real-world feasibility of such systems remains largely unexplored.

Purpose of the Study:

• To demonstrate the feasibility of a novel mobile cloud system, myBlackBox, for identifying and logging unusual personal events.
• To develop an efficient system that fuses multimodal smartphone sensor data.
• To evaluate the system's accuracy, customization, convenience, and scalability.

Main Methods:

• Developed a hybrid architecture combining unsupervised audio, accelerometer, and location data classification.
• Employed supervised joint fusion classification for enhanced accuracy.
• Implemented an end-to-end system integrating Android smartphones with cloud servers.

Main Results:

• Successfully implemented and evaluated the myBlackBox system.
• Demonstrated the feasibility of fusing multimodal sensor data for event detection.
• Achieved high accuracy, customization, convenience, and scalability.

Conclusions:

• The myBlackBox system is a feasible and practical solution for real-world mobile event logging.
• Hybrid classification approaches effectively fuse multimodal sensor data.
• The system has potential applications in personalized mobile services and context-aware computing.