Next Location Prediction Based on an Adaboost-Markov Model of Mobile Users

Area of Science:

• Data Science
• Computer Science
• Urban Planning

Background:

• Mobile user trajectory data offers insights into behavior, interests, and is crucial for smart cities and transportation planning.
• Identifying significant locations and predicting future movements from this data is challenging but valuable.
• Existing methods for trajectory analysis and prediction have limitations in accuracy and adaptability.

Purpose of the Study:

• To develop a novel method for preprocessing trajectory data to identify important user locations.
• To propose an Adaboost-Markov model for predicting the next important location of mobile users.
• To evaluate the performance and universality of the proposed prediction model.

Main Methods:

• A new trajectory data division method was proposed, extracting feature points based on structural changes.
• An improved density peak clustering algorithm was used to cluster feature points and identify important locations.
• A multi-order fusion Markov model, enhanced by the Adaboost algorithm, was developed for next-location prediction, adaptively determining model order and weights.

Main Results:

• The Adaboost-Markov model demonstrated superior prediction performance compared to a multi-order fusion Markov model with equal coefficients.
• Experimental results on the Geo-life dataset confirmed the Adaboost-Markov model's enhanced universality and prediction accuracy over first- to third-order Markov models.
• The method effectively identifies important locations and predicts future user movements.

Conclusions:

• The proposed trajectory preprocessing and Adaboost-Markov model offer a significant advancement in mobile user location prediction.
• The Adaboost algorithm effectively optimizes multi-order Markov models for improved predictive power.
• This research contributes to more accurate trajectory data analysis for applications in smart cities and personalized services.