A future location prediction method based on lightweight LSTM with hyperparamater optimization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Geopositioning data from mobile devices (GPS, GLONASS, Galileo) enables numerous location-based services.
• These datasets hold significant potential for predicting human movement and other applications.
• Existing methods may require substantial computational resources, limiting their use on edge devices.

Purpose of the Study:

• To develop a simple, lightweight machine learning method for future location prediction.
• To enable location prediction on devices with lower computing capabilities, such as Artificial Intelligence of Things (AIoT) and Edge Machine Learning (EdgeML).
• To optimize a Long Short-Term Memory (LSTM) model for efficient geopositioning data analysis.

Main Methods:

• Utilized a basic Long Short-Term Memory (LSTM) neural network model.
• Performed hyperparameter optimization, focusing on the window size for continuous geopositioning data.
• Applied the method to both continuous and non-continuous geopositioning datasets.

Main Results:

• The proposed method demonstrated considerably fast and accurate future location predictions.
• Performance was superior compared to existing neural network-model-based approaches.
• The method proved equally effective on non-continuous geopositioning data.

Conclusions:

• The developed lightweight machine learning method offers an efficient solution for future location prediction.
• The approach is suitable for deployment on resource-constrained devices like AIoT and EdgeML.
• The study validates the effectiveness of the optimized LSTM model for diverse geopositioning data types.