[Paper Review] DTNC: A New Server-side Data Cleansing Framework for Cellular Trajectory Services

Jian Dai, Fei He+3 more|arXiv (Cornell University)|Mar 1, 2017

Human Mobility and Location-Based Analysis26 references3 citations

TL;DR

DTNC proposes a server-side data cleansing framework for cellular trajectory services that leverages a dynamic transportation network (DTN) with probabilistic travel time distributions and a travel-time-aware hidden semi-Markov model (TT-HsMM) to infer accurate, complete trajectories from noisy, error-prone, and incomplete cellular location data. It reduces spatial errors below 50 meters in 40% of locations and outperforms six state-of-the-art methods in accuracy and robustness to missing data.

ABSTRACT

It is essential for the cellular network operators to provide cellular location services to meet the needs of their users and mobile applications. However, cellular locations, estimated by network-based methods at the server-side, bear with {\it high spatial errors} and {\it arbitrary missing locations}. Moreover, auxiliary sensor data at the client-side are not available to the operators. In this paper, we study the {\em cellular trajectory cleansing problem} and propose an innovative data cleansing framework, namely \underline{D}ynamic \underline{T}ransportation \underline{N}etwork based \underline{C}leansing (DTNC) to improve the quality of cellular locations delivered in online cellular trajectory services. We maintain a dynamic transportation network (DTN), which associates a network edge with a probabilistic distribution of travel times updated continuously. In addition, we devise an object motion model, namely, {\em travel-time-aware hidden semi-Markov model} ({\em TT-HsMM}), which is used to infer the most probable traveled edge sequences on DTN. To validate our ideas, we conduct a comprehensive evaluation using real-world cellular data provided by a major cellular network operator and a GPS dataset collected by smartphones as the ground truth. In the experiments, DTNC displays significant advantages over six state-of-the-art techniques.

Motivation & Objective

• Address the critical challenge of low-quality cellular trajectory data due to high spatial errors and arbitrary missing locations in real-time location services.
• Enable accurate and complete trajectory reconstruction for mobile objects using only server-side cellular data, without access to client-side sensor data.
• Develop a scalable, efficient, and robust framework that integrates real-time traffic-aware modeling into trajectory cleansing for urban mobility applications.
• Improve the reliability of cellular trajectory services for use in traffic management, urban planning, and location-based applications.

Proposed method

• Construct a dynamic transportation network (DTN) that maintains probabilistic travel time distributions on each edge, updated continuously using real-time traffic data.
• Design a travel-time-aware hidden semi-Markov model (TT-HsMM) that models object motion by inferring the most probable sequence of edges traversed on the DTN.
• Use particle-based simulation to propagate location candidates from each cellular location, guided by adaptive transition probabilities and emission probabilities derived from traffic-aware edge distributions.
• Apply a diffusion policy to distribute particles across candidate edge fragments based on directional coherence with subsequent cellular locations, improving inference accuracy.
• Select only the most probable particle paths (based on likelihood) to infer the final cleansed physical locations, ensuring robustness to noise and missing data.
• Leverage ground truth GPS data from smartphones to validate and calibrate the framework using real-world cellular data from a major network operator.

Experimental results

Research questions

• RQ1 How can we effectively reduce high spatial errors in server-side cellular location estimates without access to client-side sensor data?
• RQ2 To what extent can a dynamic transportation network with real-time travel time distributions improve trajectory cleansing accuracy?
• RQ3 How does the TT-HsMM model outperform traditional HMM and OHMM in modeling complex mobility patterns with irregular sampling?
• RQ4 Can the framework robustly handle arbitrary missing locations in trajectories while maintaining high accuracy?
• RQ5 How do different diffusion policies (even vs. directional) affect the performance of trajectory inference under varying time gaps?

Key findings

• DTNC reduces spatial errors below 50 meters in 40% of cellular locations, significantly improving accuracy compared to raw cellular data.
• The framework outperforms six state-of-the-art methods—PF, HMM, OHMM, STRS, SnapNet, and CTrack—in both accuracy and robustness to missing data.
• With a time gap of up to 110 seconds, DTNC maintains a stable distribution of Euclidean deviation, indicating resilience to missing location sequences.
• The high-quality location rate (spatial error <50m) decreases only from 47% to 41% as missing location length increases from 20s to 110s, demonstrating strong robustness.
• The directional diffusion policy improves performance for short time gaps, but the simpler even policy is adopted by default due to comparable performance and lower complexity.
• The probability difference ratio stabilizes at particle count |𝒜| = 15, indicating optimal trade-off between accuracy and computational cost.