TL;DR
This paper evaluates three major geospatial visualization tools—Mapbox, Folium, and Power BI—by applying them to a massive dataset of 85 billion vehicle detections in Istanbul. The study provides a methodological roadmap for transforming raw sensor data into actionable urban mobility insights, highlighting how coding-based solutions offer the flexibility needed for future AI-driven smart city applications.
What Problem Does This Paper Address?
Megacities like Istanbul face staggering urban mobility challenges. While location-based sensors generate mountains of data, the real bottleneck is interpretation. Traditional, "black-box" visualization platforms often lack the flexibility and reproducibility required for modern data science workflows.
The authors address the gap between raw sensor output and urban planning by exploring how different visualization frameworks handle spatial attribute data. They argue that as we move toward AI-driven decision-making, we need visualization tools that aren't just pretty pictures, but are "coding-based" environments capable of being dynamically integrated into larger analytical pipelines.
Methods
The researchers conducted a comparative evaluation of three distinct visualization ecosystems using the same underlying traffic dataset:
- Mapbox: Utilized for its robust API and "Mapbox Studio" capabilities. It excels in high-quality, custom-styled web and mobile mapping using vector tiles, which improve load times and interactivity.
- Folium: A Python library that acts as a bridge to the Leaflet.js framework. This was chosen for its "minimal code" approach, making it ideal for data scientists who want to generate interactive heatmaps and time-series visualizations directly within a Python/Pandas workflow.
- Microsoft Power BI: Representing the Business Intelligence (BI) approach, this tool was used to create interactive dashboards that allow non-technical stakeholders to filter data by year, month, or location through a user-friendly GUI.
The study also initially considered Tableau, but it was ultimately excluded due to licensing constraints—a practical reality often faced in research.
Datasets
The study utilized a massive longitudinal dataset from the Istanbul Metropolitan Municipality (IMM) Open Data Portal:
- Timeframe: 2016–2024.
- Scale: Over 85 billion vehicle detections.
- Scope: Data includes timestamps, sensor names, geographical coordinates (lat-long), and daily vehicle counts.
- Sensor Density: The number of active unique sensors fluctuated over the study period, ranging from 397 in 2016 to 251 in 2024.
Key Insights and Ideas
- Tool-Task Alignment: The "best" tool depends entirely on the end-user. Mapbox is the winner for developers building bespoke web apps; Folium is the go-to for rapid exploratory data analysis (EDA) in Python; and Power BI is superior for executive-level reporting.
- The "Pandemic Signature": The visualizations clearly captured a precipitous decline in traffic during 2020–2021, followed by a steady recovery starting in 2022. This serves as a validation of the data's sensitivity to global events.
- Geographic Disparity: Traffic density is consistently higher on the European side of Istanbul compared to the Asian side, a finding that has significant implications for infrastructure resource allocation.
- Temporal Patterns: The study confirmed that weekday traffic significantly outweighs weekend traffic, and identified specific "busiest months" that can help city planners schedule maintenance or optimize public transit.
Key Results
The paper provides a qualitative comparison (Table 1) and several quantitative findings:
- Interactivity: Mapbox and Folium provided the highest degree of spatial accuracy and custom interactivity.
- Efficiency: Folium was noted for requiring "negligible coding effort" to produce complex heatmaps.
- Anomaly Detection: By applying Z-score algorithms, the researchers were able to visualize "unusual traffic fluctuations," providing a bridge from descriptive visualization to diagnostic analytics.
- Volume: The study successfully managed and visualized a dataset representing a staggering 85 billion+ vehicle movements, proving the scalability of these modern web-based tools.
Limitations and Future Work
The authors acknowledge several hurdles:
- Data Gaps: Missing data points (null values) were present, likely due to sensor malfunctions or uneven spatial coverage. While they removed these for this study, they suggest using interpolation or more advanced statistical "imputation" in the future.
- Descriptive vs. Predictive: This paper focuses on what happened (descriptive). The authors note that the next step is predictive modeling—using this data to forecast traffic density using AI.
- Missing Variables: To truly understand Istanbul's pulse, future models need to integrate external factors like weather conditions, socioeconomic indicators, and public transport usage.
- Licensing: The exclusion of Tableau highlights the ongoing tension between proprietary software and open-source research needs.
Summary updated