[Paper Review] Mobility prediction Based on Machine Learning Algorithms

Donglin Wang, Qiuheng Zhou+3 more|arXiv (Cornell University)|Nov 12, 2021

Human Mobility and Location-Based Analysis9 references4 citations

TL;DR

This paper proposes a support vector machine (SVM)-based mobility prediction model for vehicular users in highway scenarios, using trajectory data from SUMO simulations to classify route choices. With 400 training examples, the SVM achieves a 92.2% average prediction accuracy across varying test set sizes, demonstrating strong performance for route prediction in mobile networks.

ABSTRACT

Nowadays mobile communication is growing fast in the 5G communication industry. With the increasing capacity requirements and requirements for quality of experience, mobility prediction has been widely applied to mobile communication and has becoming one of the key enablers that utilizes historical traffic information to predict future locations of traffic users, Since accurate mobility prediction can help enable efficient radio resource management, assist route planning, guide vehicle dispatching, or mitigate traffic congestion. However, mobility prediction is a challenging problem due to the complicated traffic network. In the past few years, plenty of researches have been done in this area, including Non-Machine-Learning (Non-ML)- based and Machine-Learning (ML)-based mobility prediction. In this paper, firstly we introduce the state of the art technologies for mobility prediction. Then, we selected Support Vector Machine (SVM) algorithm, the ML algorithm for practical traffic date training. Lastly, we analyse the simulation results for mobility prediction and introduce a future work plan where mobility prediction will be applied for improving mobile communication.

Motivation & Objective

• To evaluate machine learning algorithms, particularly SVM, for predicting vehicular user mobility in mobile networks.
• To develop a practical mobility prediction model using real-world-like trajectory data from a simulated highway environment.
• To assess the performance of SVM in classifying future route choices (Route1 vs. Route2) based on historical mobility patterns.
• To demonstrate the feasibility of SVM for accurate, low-latency mobility prediction applicable to 5G and intelligent transportation systems.

Proposed method

• A highway scenario with two routes, three lanes, and one direction was modeled in SUMO to simulate vehicle mobility.
• Vehicle trajectories were generated over 100 time steps, with route choices labeled as '0' (straight) or '1' (right turn).
• A linear SVM classifier was implemented using scikit-learn with a linear kernel to predict route choices based on mobility features.
• The training dataset consisted of 400 examples, and testing was performed on 10, 20, ..., 100 examples to evaluate accuracy across scales.
• The decision boundary was visualized to assess separation quality between Route1 and Route2 predictions.
• Prediction accuracy was computed as the ratio of correctly classified test instances to total test instances.

Experimental results

Research questions

• RQ1 How accurately can an SVM model predict future route choices of vehicles based on historical mobility data in a highway scenario?
• RQ2 What is the impact of varying test set size on the prediction accuracy of the SVM-based mobility prediction model?
• RQ3 How does the SVM model perform in classifying two distinct mobility patterns (straight vs. right turn) in a simulated traffic environment?
• RQ4 Can the SVM-based prediction model be effectively applied to real-time mobility management in 5G networks?

Key findings

• The SVM model achieved an average prediction accuracy of 92.2% across multiple test set sizes, with individual accuracies ranging from 86.07% to 95%.
• With 10 test examples, the model achieved 90% accuracy, and with 100 test examples, it reached 94% accuracy.
• The model demonstrated consistent performance, with most test set sizes yielding over 90% accuracy, indicating robustness.
• The linear decision boundary at Y = -1.5 effectively separated the two route classes in the feature space, supporting reliable classification.
• False predictions were observed when test data increased, with three incorrect predictions out of 100, confirming minor misclassification under higher load.
• The simulation results confirm that SVM is a viable and accurate method for mobility prediction in highway environments, suitable for integration into mobile network management.