Evaluating the impact of cultural events using machine learning and digital footfall: a case study of Bradford City of Culture 2025

In partnership with the Centre for Cultural Value, The Audience Agency and Bradford 2025 

Beyond the box office: can AI redefine cultural evaluation by measuring how culture moves a city? 

Project overview 

Cultural events are hugely important to communities, yet the UK’s arts, culture and heritage sector struggles to evidence their social, civic and economic value. This stems from the sector’s complexity: cultural activities are largely delivered by small organisations, freelancers and temporary workforces and are funded by a mix of public, private and third-sector sources. As a result, data are scattered, inconsistent and insufficiently granular. These gaps hinder the sector’s ability to develop policy, attract investment, and align local and national strategies, pointing to a need to unite various data sources to better understand the impacts of cultural events.1, 2, 3, 4  

To address this, the National Cultural Data Observatory (NCDO) was created as a shared, centralised cultural data infrastructure. It brings together cultural, social, health and economic data in one place, providing insights through tools and dashboards to support evidence- and place-based decision-making for policymakers, funders, cultural organisations and communities.2,5 To demonstrate its potential, the NCDO developed a Bradford 2025 demonstrator, a prototype showing how datasets can be connected through a single portal to provide insights from the Bradford City of Culture 2025 programme. Among the new data sources, it explores digital footfall. Unlike traditional attendance figures and ticket sales, footfall measures the number of visitors to places and serves as a proxy indicator for town centre vitality and viability6, offering a new way to evaluate cultural impact.  

This project aimed to evaluate the success of the Bradford City of Culture 2025 programme, by using machine learning to investigate changes in footfall patterns across the Bradford district, assessing both immediate effects and potential longer-term legacy outcome. In doing so, it supports Bradford 2025’s goal of pioneering innovative evaluation methods.7 The project also responds to the UK government’s call, in the AI Opportunities Action Plan8, to expand AI use to new sectors, and provides a proof of concept for how AI and digital footprints can be leveraged to strengthen cultural evaluation. 

Data and methods  

Data Source

Anonymised footfall data were provided by HUQ and collected from mobile applications for which users had explicitly consented to the collection and sharing of their data (including weather apps, games, bus time checkers and train ticketing apps)9. The data enabled the analysis of pedestrian movement and activity on a daily basis between 2019 and 2025, across five locations within the Bradford district. The locations included the whole Bradford district, the BID (Business Improvement District), the City Centre, Lister Park and Roberts Park (Figure 1). 

Accessibility description: Figures 1A and 1B contain two maps. The first map shows the boundary of the Bradford district, highlighted in orange on top of a street basemap, to indicate its geographic extent. The second map displays the other four locations within Bradford where footfall data was collected: Bradford BID, City Centre, Lister Park and Roberts Parks. Each area is shaded in a different colour on top of a street basemap, to distinguish their boundaries and surface area. 

The footfall data monitored a subsection of the population and measured the number of unique people observed within an area on a day. Individuals re-entering the same area on the day were counted once. To ensure the data reflected pedestrian activity, home residents were excluded, along with data from driving and transport apps. As the data captured only a sample of the population, they were scaled to reflect actual footfall levels and calibrated against local population statistics and census data to account for local demographics10. 

Methods

Historical footfall data were combined with contextual factors known to influence pedestrian movement, such as weather, school/bank holidays, day of the week, the COVID-19 period etc 11. Incorporating these variables allowed the model to account for seasonal patterns and routine fluctuations in pedestrian activity. 

Model development in this project followed an iterative, trial-and-error process in Python. Several predictive modelling approaches were tested, including time series forecasting (SARIMAX) and machine learning (XGBoost). These methods have previously been used in studies evaluating the impact of extraordinary events, particularly research on the COVID-19 pandemic, by comparing business-as-usual predictions with observed data11, 12, 13. Multiple approaches were trialled to identify the most suitable model for predicting footfall and evaluating the impact of a cultural programme. 

Model performance was assessed using TimeSeriesSplit cross-validation to avoid data leakage. Following model comparisons, a Random Forest Regression model was selected as the final approach. The model was refined through hyperparameter tuning, and SHAP feature importance analysis was conducted to identify the main drivers of footfall. It was trained on 2019-2024 footfall data, and used to predict what footfall in 2025 would have looked like in the absence of the cultural programme in locations across Bradford. These predictions were then compared with observed 2025 footfall to estimate the additional activity potentially generated by the cultural events and evaluate their impact11. 

Model development faced many limitations. The dataset was relatively small and measured footfall on a daily rather than hourly basis, reducing temporal precision. Spatial resolution was coarse, covering a limited number of areas, and the model struggled to generalise across locations with distinct footfall patterns (for example between high-street areas and parks). Consequently, findings should be interpreted cautiously.  

Key findings  

Exploratory data analysis and interactive visualizations provided insights into footfall patterns across all five locations in Bradford (Figure 2). Year-on-year comparisons between 2024 and 2025 revealed significant annual uplift across the overall district (+42.3%), with variation by site. Bradford City Centre recorded the greatest localised growth, with a 12.4% annual footfall increase, representing over 4 million additional visitors. Bradford BID (+9.8%) and Roberts Park (+10.9%) also experienced a rise in footfall, while Lister Park saw a slight decline (-2.5%). Monthly analysis highlighted February as the peak growth period in the City Centre (+32.4%), coinciding with the installation and unveiling of the Ramadan Pavillion structure, suggesting an association between the Bradford City of Culture 2025 programme and increased pedestrian activity. 

Accessibility description: Figure 2 is a screenshot of an interactive dashboard displaying a line chart. A dropdown menu is set to Bradford City Centre, indicating that this area has been selected for analysis. The chart shows daily footfall counts on the vertical (y) axis and the months of the year, from January to December, on the horizontal (x) axis. Two lines are plotted on the same graph: a blue line representing footfall in 2024 and a red line representing footfall in 2025. The lines allow comparison of daily trends across the two years. In the top-right corner of the chart, a summary statistic reports the overall annual percentage change in footfall between the two years for this area, showing an increase of 12.4%. 

However, simple year-on-year comparisons risk overstating impact, as they fail to account for external factors influencing pedestrian activity such as adverse weather conditions or construction activity. The Random Forest Regression model was developed as a proof of concept to demonstrate how machine learning can enhance cultural impact evaluation. Results from the model are by no means definitive, however the potential for measuring changes in footfall patterns attributable to the cultural programme exists. The model achieved good predictive performance (R2= 0.97, MAE=26871 people/day, RMSE= 41038 people/day, MAPE= 0.52), allowing to predict normal footfall patterns based on contextual factors. Comparing predicted baseline footfall (i.e. expected levels without cultural events) with observed 2025 figures provided an adjusted estimate of programme impact. 

In the City Centre, annual footfall was 6.8% higher than expected in 2025, which is a slightly lower uplift compared to the naïve 12.4% year-on-year increase (Figure 3). At the district level, observed annual footfall exceeded modelled expectations by 34.9%, while Bradford BID showed minimal difference between the year-on-year comparison and modelled estimates (+9.8 vs +9.5%). In contrast, both park locations recorded lower than expected annual footfall relative to model predictions (Lister Park -18.4%, Roberts Park -57.4%), suggesting that observed changes in these areas were more strongly influenced by contextual dynamics than by the cultural programme. Feature importance analysis (Figure 4) identified location and whether a day fell within the COVID-19 period as the strongest predictors of footfall, reinforcing the importance of contextual variables in pedestrian movement modelling. 

Accessibility description: Figure 3 is a line chart showing daily footfall counts across the 2025 year for the Bradford City Centre. The horizontal (x) axis displays the months from January to December. The left vertical (y) axis represents the daily footfall counts. Three lines are plotted on the chart: a red line representing the model-predicted footfall, a black line representing observed footfall, and a purple dashed line representing the percentage difference between the predicted and actual values. A second vertical axis on the right side of the chart displays the percentage difference scale. Together, these lines allow comparison between predicted and observed daily trends throughout the 2025 year. In the bottom-right corner of the chart, a summary statistic reports the overall annual percentage change between predicted and observed footfall, indicating an increase of 6.8%. 

Accessibility description: Figure 4 is a horizontal bar chart showing the feature importance of the model, measured using mean absolute SHAP values. The horizontal (x) axis represents the mean absolute SHAP values, indicating the overall contribution of each feature to the model’s predictions. The vertical (y) axis lists the features, arranged from most important at the top to least important at the bottom. Each feature is displayed as a horizontal bar whose length corresponds to its importance. The bars are colour-coded according to feature category. A legend in the lower-right corner identifies the categories and their associated colour: calendar, external shock, temporal, weather and location. The most important features are the location variables, which occupy the top positions in the ranking. The next most important feature outside of the location category is the COVID-19 variable, which appears as the sixth most important feature. 

Overall, given the data quality and coverage limitations, these findings should not be interpreted as definitive measures of programme impact. Rather, the project demonstrates the potential of AI approaches to enhance data-informed cultural evaluation. With improved data, including better temporal and spatial resolution, future research could refine the model. This would enable to move beyond simple descriptive measures of cultural impact towards predictive modelling that estimates change relative to expected conditions, supporting a more robust and credible assessment of impact. 

Value of the research  

This project serves as a proof of concept for a more innovative, data-driven approach to cultural evaluation. It demonstrates how integrating new datasets and applying AI and data science methods can improve our understanding of how cultural events contribute to neighbourhood vitality, moving away from traditional economic impact measures. The research supported the development of the NCDO, generated insights for the Bradford 2025 team, and fostered collaboration across many academic, industry and cultural partners. It also highlighted critical data gaps within the sector and laid the foundation work for future research and cultural evaluation, illustrating how richer and more granular data could enable better place-based and policy-relevant cultural evaluation. 

Quotes from project partner 

“Our data scientist played a pivotal role in translating complex footfall and contextual datasets into a robust, transparent model capable of estimating cultural impact in real-world conditions. By combining methodological rigour with collaborative working across academic, civic and industry partners, this work demonstrates how data science can strengthen evidence for cultural value in ways that are both innovative and policy-relevant.” 

Stephen Dobson, Director of Centre for Cultural Value, University of Leeds 

"It has been great to work with a LIDA data scientist on this project, using AI and Machine Learning to help us understand the impact Bradford 2025 City of Culture has had across the Bradford district. The project has shown how these methods can be used to analyse large datasets and uncover valuable insights about engagement across our communities. It’s a great example of how AI can support cultural organisations to better evidence and understand their impact." 

Amel Hadfi, Data Analyst, Bradford 2025 

Insights 

• Developed a novel, data-driven framework for cultural evaluation, using digital footfall data as proxy for neighbourhood vitality. 
• Applied AI and machine learning methods, previously used in COVID-19 research, in a field that has rarely used data science approaches.  
• Established lasting cross-sector collaboration between academic, industry and cultural stakeholders. 
• Identified critical gaps in data availability, granularity and standardisation that currently limit robust cultural impact assessment. 
• Provided a scalable proof of concept that can be strengthened and reused as richer, more granular datasets become available in the future. 

Team 

• Marion Carneiro, Data Scientist, Leeds Institute for Data Analytics, University of Leeds 
• Stephen Dobson, Director of Centre for Cultural Value, University of Leeds 
• Stephen Miller, Chief Technology Officer, The Audience Agency 
• Amel Hadfi, Data Analyst, Bradford 2025 
• Nick Malleson, Professor of Spatial Science, University of Leeds 

Partners

• Centre for Cultural Value 
• The Audience Agency 
• Bradford 2025 

Funder 

Funded by the Centre for Cultural Value and University of Leeds ESRC Impact Acceleration Award funding ring-fenced for LIDA’s Data Scientist Development Programme. 

This work has been facilitated by the Leeds Institute for Data Analytics (LIDA) Data Scientist Development Programme, which employs early-career data scientists to deliver real-world data-driven impact in the interests of the public good. 

References 

1. Centre for Cultural Value. 2025. Making data work. [Online]. [Accessed 20 February 2026]. 
2. Elliott, J. 2025. Next steps in transforming cultural sector data, Centre for Cultural Value. [Online]. [Accessed 20 February 2026]. 
3. Department for Culture, Media and Sport. 2022. Better data on the Cultural Economy: Scoping Study. [Online]. [Accessed 20 February 2026].
4. World Cities Culture Forum. 2025. The power of Cultural Data. [Online]. [Accessed 20 February 2026].
5. National Cultural Data Observatory. 2025. Manifesto. [Online]. [Accessed 20 February 2026]. 
6. Mumford, C. et al. 2020. Footfall signatures and volumes: Towards a classification of UK Centres. Environment and Planning B: Urban Analytics and City Science. 48(6), pp. 1495–1510. doi:10.1177/2399808320911412.  
7. Bradford 2025. 2024. Bradford 2025 UK City of Culture evaluation strategy 2023-2026 version 1. [Online]. [Accessed 20 February 2026].
8. Department for Science, Innovation & Technology. 2025. AI Opportunities Action Plan. [Online]. [Accessed 20 February 2026].
9. Huq Industries. No date. How many people visit my location? [Online]. [Accessed 20 February 2026].
10. Huq Industries. No date. FAQs & support documentation. [Online]. [Accessed 20 February 2026].
11. Asher, M., Oswald, Y. and Malleson, N. 2025. Understanding pedestrian dynamics using machine learning with real-time urban sensors. Environment and Planning B: Urban Analytics and City Science. 52(8), pp. 1994–2017. doi:10.1177/23998083251319058.  
12. Seyidoglu, H. et al. 2024. Post-pandemic crime trends in England and Wales, Crime Science. 13(1). doi:10.1186/s40163-024-00201-1.  
13. Enoch, M. et al. 2021. When covid-19 came to town: Measuring the impact of the coronavirus pandemic on footfall on Six High Streets in England. Environment and Planning B: Urban Analytics and City Science. 49(3), pp. 1091–1111. doi:10.1177/23998083211048497.