# Zen: LSTM-based generation of individual spatiotemporal cellular traffic with interactions

**Authors:** Anne Josiane Kouam, Aline Carneiro Viana, Alain Tchana  
**arXiv:** [2301.02059](https://arxiv.org/abs/2301.02059)  
**Date:** 2023-01-05  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2301.02059v1/

## Key points

- Zen is a novel framework designed to address accessibility, usability, and privacy issues in cellular network datasets, specifically focusing on Charging Data Records (CdRs).
- Traffic Behavior Modeling: Uses Long-Short-Term Memory neural networks to capture long-range correlations and inter-CdRs specificity with high accuracy in modeling event types and inter-event durations.
- Mobility Behavior Emulation: Emulates individual mobility behaviors in a real-world metropolitan city setting to enhance the emulation of urban daily-life mobility behaviors.
- Cellular Network Organization: Reproduces cellular network organization with multiple operators and establishes social ties between users for flexibility in producing CdRs for various operators simultaneously.
- Comprehensive CdRs Generation: Generates complete CdRs describing individual mobility, traffic patterns, and pairwise communications following real traffic behavior.
- Zen's ability to accurately capture individual and global distributions of real-world CdRs datasets is validated through various experiments.
- Zen demonstrates utility in practical networking applications such as dynamic population tracing, Radio Access Network power savings, and anomaly detection compared to real-world CdRs.

## Summary

This paper introduces Zen, a novel framework designed to address the challenges associated with accessibility, usability, and privacy issues in cellular network datasets. Specifically focusing on Charging Data Records (CdRs), Zen aims to enhance research reproducibility and enable more effective exploitation of these datasets by modeling real-world data attributes. The framework follows a four-fold methodology
1. Traffic Behavior Modeling: Utilizing a fully anonymized CdRs dataset describing users' traffic behavior, Zen employs Long-Short-Term Memory neural networks to capture long-range correlations and inter-CdRs specificity while addressing population heterogeneity. The model achieves high performance values with significant accuracy in modeling event types and inter-event durations. 2. Mobility Behavior Emulation: By emulating individual mobility behaviors in a real-world metropolitan city setting (Helsinki EU city), Zen incorporates infrastructure and cell tower distribution information to enhance the emulation of urban daily-life mobility behaviors. 3. Cellular Network Organization: A separate module is designed to realistically reproduce cellular network organization with multiple operators and establish social ties between users, enabling flexibility in producing CdRs for various operators simultaneously. 4. Comprehensive CdRs Generation: By combining the previous models, Zen generates complete CdRs that describe individual mobility, traffic patterns, and pairwise communications following real traffic behavior. The framework's ability to accurately capture individual and global distributions of real-world CdRs datasets is validated through various experiments. Additionally, Zen demonstrates its utility in practical networking applications such as dynamic population tracing, Radio Access Network power savings, and anomaly detection compared to real-world CdRs. Overall, represents a significant advancement in generating realistic  on an individual basis, offering new insights into human presence and activity studies while preserving . Through the modeling of realistic emulated mobility behaviors alongside real-world traffic data, Zen provides a comprehensive solution to the challenges associated with , , and  in cellular network datasets.