Adaptive activity and environment recognition for mobile phones

Area of Science:

• Computer Science
• Mobile Computing
• Machine Learning

Background:

• Activity and environment recognition on mobile devices is crucial for context-aware applications.
• Existing methods often lack adaptability to individual user characteristics and real-world variability.

Purpose of the Study:

• To develop and evaluate an adaptive activity and environment recognition algorithm for mobile phones.
• To enhance classification accuracy through user-specific adaptation and a confidence measure.

Main Methods:

• Utilized sensor and radio receiver data from mobile phones.
• Employed a Bayesian maximum a posteriori (MAP) classifier, Support Vector Machines (SVM), and Decision Trees (DT).
• Developed an adaptation algorithm for context model parameters and a confidence measure for classification correctness.

Main Results:

• The proposed algorithm demonstrated accurate classification of user activities and environments.
• The adaptive algorithm significantly improved classification accuracy in real-life trials.
• Comparison showed competitive or superior performance against SVM and DT classifiers.

Conclusions:

• The presented adaptive algorithm effectively enhances mobile-based activity and environment recognition.
• The confidence measure provides valuable insights into classification reliability.
• This approach offers a robust solution for personalized context-aware mobile systems.