Study of LZ-based location prediction and its application to transportation recommender systems

Area of Science:

• Computer Science
• Ubiquitous Computing
• Mobile Computing

Background:

• Anticipating user context through next location prediction is crucial for timely reactions.
• Existing methods often require significant computational resources, limiting their use on mobile devices.

Purpose of the Study:

• To evaluate LZ-based algorithms (LZ, LeZi Update, Active LeZi) for efficient next location prediction on mobile devices.
• To explore combinations of algorithm phases for optimizing prediction accuracy and resource consumption.

Main Methods:

• Utilizing a dataset of GSM-based mobility traces from 95 users.
• Dividing LZ-based algorithms into two phases for flexible combination and analysis.
• Comparing different algorithm phase combinations to assess prediction performance and resource usage.

Main Results:

• LZ-based algorithms demonstrate low resource requirements suitable for mobile execution.
• Specific combinations of algorithm phases yield improved prediction accuracy and reduced resource consumption.
• The study validates the practical application of these predictors in real-world scenarios.

Conclusions:

• LZ-based algorithms offer a viable solution for on-device next location prediction.
• Optimized algorithm phase combinations can enhance the efficiency and accuracy of mobility pattern learning.
• The developed prototype showcases the utility of location prediction in ubiquitous computing environments, such as public transport recommendation systems.