eXtended Detail Records (XDRs) in Telecom

• eXtended Detail Records (XDRs) are comprehensive event-level logs that capture both human and device-initiated interactions across mobile networks.
• They facilitate advanced mobility and home location detection through high temporal resolution and nuanced event categorization.
• Analytical techniques like clustering and predictive modeling applied to XDRs enhance anomaly detection and urban accessibility studies in telecom research.

eXtended Detail Records (XDRs) are comprehensive event-level logs generated by mobile communication systems, detailing both user- and device-initiated interactions across data networks. Distinguished from Call Detail Records (CDRs), which are limited to voice and SMS events typically triggered by explicit user actions, XDRs encapsulate a wider scope of transactions, including system-level, application-level, and periodic device events, thereby offering a richer temporal and behavioral resolution for studies in telecommunications, mobility, and network analytics.

1. Structure and Generation Mechanisms of XDRs

XDRs capture events arising from both human activity (such as web browsing, app use, email retrieval) and automated device operations (including background data synchronization, periodic system heartbeats). Unlike the strictly human-triggered CDRs and the network-internal Control Plane Records (CPRs), which register machine-to-network signaling, XDRs straddle these categories: generated for each billable data event or relevant network-side incident, they provide a temporally fine-grained record array that is less sparse than CDRs yet more semantically linked to actual mobility and service usage than CPRs (Pappalardo et al., 2020).

This granularity is evident in high-frequency data points per user, representing not only explicit interaction (such as starting a video stream) but also implicit, network-background operations, which result in a multifaceted behavioral footprint.

2. Role in Mobility and Home Location Detection

XDRs have been foundational in a new generation of mobility analytics. Their high temporal and spatial density facilitates finer-grained inferences of user trajectories, home locations, and work sites than is possible with CDRs. For home location detection, XDRs are processed using Home Detection Algorithms (HDAs) that aggregate records by mobile tower, adjusting for event type and time window.

A key methodological advance leverages the "hour-of-day" HDA (HDA3), which identifies the home antenna as the one receiving maximal nighttime activity (typically 19:00–07:00) (Pappalardo et al., 2020, Marin-Flores et al., 29 Jul 2025). This approach exploits both the regularity of device-generated background events and the persistence of user presence at night. Empirical results demonstrate that, when using XDRs, HDA3 attains an accuracy of approximately 68% in identifying the correct home tower (within the three nearest to the actual residence), outperforming CDRs and providing a superior compromise between event volume and behavioral relevance.

A summary of the data streams is presented below:

Data Stream	Event Trigger	Temporal Granularity
CDRs	Human-initiated (calls, SMS)	Sparse
XDRs	Human and device (data use, sync)	Moderate-High
CPRs	Network-initiated (signaling)	Extremely High

3. Analytical Techniques: Clustering, Prediction, and Statistical Modeling

XDRs' extended feature sets enable advanced analytics. Unsupervised clustering, such as k-means, has been used to isolate network anomalies by partitioning event records spatially and temporally. Anomalous behavior is identified via small, outlying clusters—which may signal events such as natural disasters, network faults, or security incidents (Sultan et al., 2018). This process involves initializing cluster centers, assigning XDR points to the nearest centroid, and iteratively updating centroids until convergence. The cluster with minimal membership is flagged as anomalous and cross-validated against ground truth via temporal plots.

For predictive modeling, XDRs provide enhanced input for neural network training and time series forecasting. Neural networks trained on XDRs that have been cleaned of anomalies yield lower mean square error (MSE), both in-sample (training) and out-of-sample (validation), reflecting a superior generalization capability. Furthermore, ARIMA models for traffic forecasting capitalize on the statistical richness of XDRs, incorporating autoregressive, integrated (differenced), and moving average terms:

• AR: $$Y_t = c + \sum_{i=1}^p \phi_i Y_{t-i} + \varepsilon_t$$
• I: $Y'_t = Y_t - Y_{t-1}$
• MA: $$Y_t = \mu + \varepsilon_t + \sum_{i=1}^q \theta_i \varepsilon_{t-i}$$

Stationarity is ensured via differencing and the Augmented Dickey-Fuller (ADF) test prior to model fitting. Preprocessing XDRs to extract anomaly-free subsequences consistently yields stronger predictive performance.

4. Applications in Urban Mobility and Accessibility Research

XDRs enable large-scale dynamic analysis of urban mobility patterns, obviating limitations of surveys and static census data (Marin-Flores et al., 29 Jul 2025). For instance, in Santiago, Chile, XDRs covering nearly one-third of the mobile subscriber base have been used to algorithmically determine home and work locations by weighted aggregation of BTS connections—home locations from nighttime signals (hours weighted by nocturnality; e.g., Wh(2,3)=3, Wh(0,1,4,5)=2, Wh(6–23)=1; zero otherwise) and work locations similarly partitioned by working hours.

Subsequent accessibility analyses employ the R5 multimodal routing engine to model travel times and multimodal connectivity (using GTFS and OSM data). Socioeconomic disparities are examined through metrics such as the Palma ratio and the Hansen cumulative opportunities index:

$COA_o = \sum_{d} P_d \cdot f(t_{od})$

where $COA_o$ denotes cumulative opportunities for origin $o$, $P_d$ the opportunity count at destination $d$, and $f(t_{od})$ the impedance function based on travel time.

Spatial patterns are further interrogated via bivariate Local Indicators of Spatial Association (LISA), revealing clusters where social-material disadvantage, commuting times, and demographic disproportions (e.g., indigenous population prevalence, gender disparities) converge. This analytical schema establishes XDRs as an indispensable data source for describing—and statistically quantifying—urban accessibility and social inequality.

5. Data Minimization and Privacy Implications

XDRs' denser event streams afford effective home or mobility inference with fewer total records than CDRs, important for both computational efficiency and privacy. In calibration studies, as little as a moderate sample of XDRs suffices for robust inference, contrasting with sparser CDRs (which necessitate a larger percentage of the available data) and outstripping CPRs (which may require as little as 10%, but can lose behavioral signal in the bulk of machine-triggered noise) (Pappalardo et al., 2020).

For researchers coordinating with telecom providers, these characteristics enable more targeted data requests—minimizing both required data volume and exposure of sensitive behaviors, while maintaining high accuracy for critical applications such as home detection or anomaly localization.

6. Methodological Limitations and Prospects

While XDRs represent a methodological advance, their interpretation for behavioral inference requires careful algorithmic design. The accuracy of any inference—home detection, anomaly detection, or mobility estimation—depends strongly on the filter and aggregation schemes (e.g., hour-of-day windows, event weighting), properties of the dataset (coverage, sampling interval), and the social context of device use.

The literature demonstrates that algorithmic choices (e.g., the superiority of HDA3 for XDR-based home inference) are not invariant across user populations or markets. A plausible implication is that cross-validation with ground truth, systematic benchmarking of detection algorithms, and context-aware parameter tuning remain essential for high-confidence inference.

The application of XDRs is expanding in precision urban analytics, real-time network monitoring, and integrated socio-technical systems research. Their adaptability to established machine learning and time series frameworks—combined with their event richness and intermediate balance between behavioral and system triggers—positions them as a central data stream for the operationalization of mobility, anomaly detection, and accessibility analytics at population scale.