Mobile phone networks have quietly become one of the most powerful scientific instruments for understanding how cities breathe, and a new study from Italy demonstrates just how much can be learned when millions of anonymized movement records are put under the microscope. In research published in the Journal of Ambient Intelligence and Humanized Computing, a team at the University of Modena and Reggio Emilia, working with ENEA, has built a complete data-driven pipeline that characterizes human mobility across the Emilia–Romagna region, automatically detects large public events, forecasts tourist inflows with around 10 percent error at regional scale, and packages the results into a working decision-support application now being used by a real municipality.
The dataset at the heart of the study is remarkable in both scale and granularity. Provided by the mobile operator TIM Enterprise, it covers 60 consecutive days in August and September 2019 and reports millions of daily origin–destination flows between Elementary Census Areas, the smallest administrative units in the region, each one enriched with demographic and behavioral attributes. Every record captures how many people moved from one census area to another during a given interval, decomposed into six age cohorts, male and female counts, Italian and foreign citizens, and behavioral labels distinguishing business-related trips from consumer and leisure travel. These attributes are derived from SIM card and contract registration data, while trip purpose is inferred by the operator using proprietary algorithms based on mobility patterns and network usage, not from surveys.
Because raw operator data is subject to privacy-preserving suppression protocols required by the GDPR, many low-volume records contain demographic sub-categories that have been masked or zeroed out. The team therefore applied a stringent multi-step filtering process. They first restricted the analysis to intra-regional movements, then enforced temporal consistency checks that discarded records with negative durations or trips lasting more than an hour, and finally imposed a strict demographic coherence filter, retaining a flow only if the sum of its sub-components across age, gender, nationality, and trip purpose exactly matched the total. This rigorous cleaning reduced the dataset from an initial 36,309,517 raw records to 22,962,575 robust, fully consistent records, discarding more than 13 million heavily censored micro-movements and yielding a dataset of exceptional reliability.
The choice of August and September was deliberate rather than incidental. The two months capture two fundamentally different mobility regimes: August reflects the Italian holiday period, with dispersed, leisure-driven travel, while September marks the resumption of work commutes and the school year. Split violin plots comparing daily mobility distributions between the months revealed exactly this contrast, with August exhibiting lower and highly variable volumes and September displaying higher, tightly concentrated movement patterns. The aggregate time series showed clear weekly seasonality, with pronounced weekend dips, particularly on Sundays, and a gradual upward trend as summer ended. Hourly patterns proved remarkably stable, however: roughly 35 to 40 percent of daily movements consistently occurred between noon and 5 p.m., regardless of the month.
Demographic disaggregation added further texture. People over 60 were consistently the most mobile group during the observation window, likely reflecting seasonal tourism, while minors moved the least. Mobility among foreign citizens remained stable through August before declining in September, consistent with post-summer departures. At the city scale, a case study of Bologna identified the ten most intense internal flows, which clustered around the historic center, the Central Railway Station, and Piazza Maggiore, with strong corridors extending along Via San Donato, Via Andrea Costa, and Via Emilia Levante. Flows between Bologna and its surrounding province radiated outward from the city, confirming its role as the region’s dominant mobility hub, with key connections to Calderara di Reno, Castel Maggiore, San Lazzaro di Savena, and Granarolo dell’Emilia. On weekdays, longer-distance flows concentrated along functional corridors to Modena, Ferrara, and the Romagna coast, while weekend flows spread more evenly toward coastal and recreational areas. During August, the network’s topology remained structurally identical, but absolute volumes on top commuting routes dropped by more than 30 percent.
Beyond description, the researchers built a predictive framework to forecast daily inflows into individual census areas. Feature engineering captured weekly and seasonal cycles through temporal identifiers such as weekday, week of year, and month, along with binary weekend and holiday flags, autoregressive lag features representing arrivals on previous days, and exogenous variables including weather events and localized cultural or sporting events. Regional proxies, namely aggregate inflows from the neighboring hubs of Bologna and Imola recorded the previous day, allowed the model to account for mobility pressure propagating across the network. Target values were log-transformed to handle the strong right-skew of flow distributions, and a Gradient Boosting Regressor, optimized via grid search, was trained on 70 percent of the data and tested on the remainder. Urban and industrial hubs such as Bologna and Modena proved the most predictable, with errors of roughly 10 to 15 percent attributable to the stability of commuting patterns, while tourist-centric and mountainous areas showed errors below 25 percent, remaining within bounds suitable for planning purposes even in highly variable zones.
Perhaps the most striking demonstration, however, is the automated event detection. For each census area and hour, the team computed a Z-score comparing the observed net flow against the historical mean and standard deviation for that location. Values exceeding three standard deviations in either direction were flagged as potential anomalies. The method worked spectacularly well where the signal-to-noise ratio was high. At the census area containing the Bologna Exhibition Centre, a pronounced anomaly from 23 to 27 September aligned precisely with the opening of Cersaie, the international ceramics fair that draws tens of thousands of visitors, while a weaker peak in early September corresponded to the SANA organic products exhibition. At the Imola Autodrome, anomalies on 7 September matched the CRAME Swap Meet, one of Europe’s largest vintage vehicle fairs, complete with the characteristic outbound spike at 5 p.m. as crowds dispersed, and a further cluster from 20 to 22 September captured the Summer Food Experience. In Modena, Z-scores exceeding 3.5 tracked the start of the school year, the Festivalfilosofia festival, and the Modena Motor Gallery.
The study is unusually honest about the method’s limits. Dedicated event venues show near-zero baseline mobility on quiet days, producing a low standard deviation and a very high signal-to-noise ratio, so events emerge cleanly. Mixed-use urban centers, by contrast, carry heavy routine traffic from commuters, students, and residents, inflating the historical standard deviation and dampening Z-scores, which caused mid-sized events like the SANA exhibition to barely reach the detection threshold. Conversely, urban areas are prone to false positives: a massive anomaly in Modena on 30 September, with a Z-score of 5.91, turned out to be ordinary end-of-month congestion combined with weather-related disruptions rather than any public event. The authors propose context-aware baselines, comparing, for instance, a Tuesday evening only against historical Tuesday evenings, as a future remedy.
The final phase of the work zoomed in on Dozza, a small historic tourist town near Bologna whose limited number of access points makes it ideal for systematic monitoring. Smart cameras installed at three locations recorded hourly pedestrian entries and exits from March 2022 to September 2025, with measurements correlating at better than 0.95 across cameras. Entries peaked around midday and exits around 5 p.m., with weekends showing substantially higher footfall. The team fused this ground truth with a TIM presence dataset that classifies mobile users as tourists, excursionists, transients, or residents. Only the excursionist category correlated significantly with camera counts. Crucially, the analysis revealed that pedestrian peaks in Dozza were often driven by events beyond the municipality itself, including the Formula 1 Grand Prix at Imola, major trade fairs such as MECSPE, COSMOPROF, and CERSAIE in Bologna, and activity surges at Bologna Airport, effects the researchers captured through simple lagged binary indicators marking weekends following high-activity weeks.
Two complementary Gradient Boosting experiments, evaluated with leave-one-month-out cross-validation, showed that the mobility-based feature set achieved better absolute accuracy, with an RMSE of 218.26 and MAE of 152.02, while the event-based set achieved a lower MAPE of 33.40 percent and, critically, predicted the exact dates of pedestrian peaks more reliably. A probabilistic classifier identifying peak days above 500 pedestrians flagged at least one probability peak in every observed high-attendance period. These models are now embedded in an operational decision-support application built for the municipality, structured around a backend service, a web frontend, a PostgreSQL database, and a pedestrian prediction service, allowing administrators to define what-if scenarios for parking and public transport and to see the estimated infrastructure load under low-event and high-event conditions. The researchers argue that the framework offers a dual-purpose tool, supporting both regional transport planning and fine-grained tourism management, and that future work will extend it to real-time data sources and multi-modal, regional-scale scenarios, bringing data-driven sustainability a step closer to everyday urban governance.
Subject of Research: Large-scale analysis, anomaly detection, and forecasting of human mobility and tourist inflows in Emilia–Romagna, Italy, using anonymized mobile network origin–destination data.
Subject of Research: Technology and Engineering
Article Title: From flows to events: spatiotemporal insights into Bologna’s mobility using mobile networks
Article References: Bicocchi, N., Arioli, M., Mamei, M., & Petrovich, C. (2026). From flows to events: spatiotemporal insights into Bologna’s mobility using mobile networks. Journal of Ambient Intelligence and Humanized Computing. https://doi.org/10.1007/s12652-026-05127-x
Image Credits: AI Generated
DOI: 10.1007/s12652-026-05127-x
Keywords: urban mobility, mobile network data, anomaly detection, machine learning, Gradient Boosting, smart cities, origin–destination flows, tourist inflow prediction, Z-score, event detection, decision support, Emilia–Romagna
Cite Scienmag News
APA
MLA
Chicago
Denise Maddox. (September 11, 2026). Mobile network data reveals Bologna’s mobility patterns through spatiotemporal event analysis. Scienmag. https://scienmag.com/mobile-network-data-reveals-bolognas-mobility-patterns-through-spatiotemporal-event-analysis/
Denise Maddox. “Mobile network data reveals Bologna’s mobility patterns through spatiotemporal event analysis.” Scienmag, 11 September 2026, https://scienmag.com/mobile-network-data-reveals-bolognas-mobility-patterns-through-spatiotemporal-event-analysis/. Accessed 11 September 2026.
Denise Maddox. “Mobile network data reveals Bologna’s mobility patterns through spatiotemporal event analysis.” Scienmag. September 11, 2026. https://scienmag.com/mobile-network-data-reveals-bolognas-mobility-patterns-through-spatiotemporal-event-analysis/
Copy citation
Download RIS
Tags: anonymized mobile phone dataanonymized movement datadecision-support applications for citiesdemographic analysis of mobilityEmilia–Romagna mobility researchEmilia–Romagna region mobility studyhuman mobility characterizationhuman movement forecastinglarge-scale demographic and behavioral datamobile data-driven city planningMobile network data analysismobile phone network as scientific instrumentpublic event detection using mobile datapublic event impact on mobilityregional scale mobility studiesregional tourism inflow predictionregional tourist inflow forecastingspatiotemporal event detectionurban mobility patterns
