Enhancing Sustainable Urban Mobility Prediction with Telecom Data: A Spatio-Temporal Framework Approach

TL;DR

This work tackles the lack of directionality in telecom-derived traffic data by introducing TeltoMob, a dataset combining undirected GCT counts with directed mobility flows, and a two-stage STGNN framework to infer directional mobility. Stage 1 pre-trains a STGNN on GCT flows to extract rich spatio-temporal features $H\in\mathbb{R}^{N\times C\times D}$, while Stage 2 transforms these features into route-aligned representations ${h}_{\overline{ij}}$ via ${h}_{\overline{ij}}=\sigma(h_j-h_i)$, enhances them with 1-hop upstream context through MGAT, and predicts mobility with a second STGNN followed by an MLP. Empirical results show consistent improvements across multiple STGNN baselines (up to $17.5\%$ RMSE improvement for longer horizons) and demonstrate the framework’s potential for real-time deployment as a traffic indicator to support sustainable urban mobility. The dataset and framework together offer a cost-effective path to leverage telecom data for directional traffic understanding and control, reducing dependence on extensive physical detectors.

Abstract

Traditional traffic prediction, limited by the scope of sensor data, falls short in comprehensive traffic management. Mobile networks offer a promising alternative using network activity counts, but these lack crucial directionality. Thus, we present the TeltoMob dataset, featuring undirected telecom counts and corresponding directional flows, to predict directional mobility flows on roadways. To address this, we propose a two-stage spatio-temporal graph neural network (STGNN) framework. The first stage uses a pre-trained STGNN to process telecom data, while the second stage integrates directional and geographic insights for accurate prediction. Our experiments demonstrate the framework's compatibility with various STGNN models and confirm its effectiveness. We also show how to incorporate the framework into real-world transportation systems, enhancing sustainable urban mobility.

Figures (8)

• Figure 1: Overview of the task and framework. Network activities collected at road areas (points 1 to 3) act as proxies for traffic conditions but lack crucial directionality for accurate traffic management. Our framework leverages non-directional telecom data from past time steps to predict future directional mobility flows, enhancing its utility for urban computing.
• Figure 2: Overview of data collected from 34 road segments, including 84 directional routes in Hsinchu City. (a) The map depicts GCTs sourced from user activity, while mobility (i.e., GCT pairing) is determined by associating GCT records appearing in adjacent segments along routes. Color intensity represents the average volumes of GCT and mobility flows. (b) Sample daily GCT flow pattern. (c) Sample daily mobility flow pattern.
• Figure 3: Histograms showing average GCT and mobility flow distribution in our dataset. The x-axis indicates flow intervals, and the y-axis counts road segments and routes. The right-skewed distribution highlights low traffic on most routes, with a few experiencing high volumes, typical of urban road network hierarchies.
• Figure 4: Relationships between GCT and mobility flows. (a) Weekly patterns show morning peaks in mobility flow for work commutes, with less evening traffic, unlike GCT flow, which reflects all directional activities. (b) Although correlations are generally positive, reduced evening mobility compared to persistent high GCT flows leads to lower correlations (indicated by blue ovals).
• Figure 5: Pearson correlation analysis of daily mobility flow for route $\overline{5\,4}$ and its upstream routes on 2022/09/05 reveals strong correlations with 1-hop upstream routes and weaker ones with 2-hop upstream routes, indicating a diminishing impact from distant routes. This finding directs our framework's emphasis on 1-hop upstream route correlations for understanding traffic continuity.