In a paper published in the journal Applied Sciences, researchers employed machine learning (ML) techniques to predict pedestrian compliance at crosswalks in urban settings across Jordan, aiming to enhance pedestrian safety and traffic management.
Study: Predicting Pedestrian Compliance Using ML. Image Credit: ultramansk/Shutterstock
Analyzing data from 2437 pedestrians at signalized intersections in Amman, Irbid, and Zarqa, the research developed and tested four ML models: artificial neural network (ANN), support vector machine (SVM), decision tree (ID3), and random forest (RF).
Results demonstrated that local infrastructure and traffic conditions significantly influenced pedestrian behavior, with the RF model proving particularly effective due to its high accuracy and precision. The study offered insights into demographic and spatial factors affecting pedestrian compliance, emphasizing ML algorithms' potential to enhance urban traffic dynamics and dynamic pedestrian safety.
Related Work
Previous global research has extensively utilized tools, software, statistical methods, and ML algorithms to analyze traffic data in urban areas. These technologies facilitate efficient data collection and analysis, providing insights into factors influencing pedestrian safety and behavior at urban intersections.
ML techniques like ANNs, CNNs, PCA, RF, SVMs, and reinforcement learning (RL) have proven effective in predicting and improving pedestrian safety and compliance at crosswalks in various studies worldwide. Recent advancements include long short-term memory (LSTM) models for forecasting pedestrian crossings during red lights, showcasing ML's potential in optimizing urban traffic dynamics and enhancing pedestrian safety.
Urban Signalized Intersections Study
Data for this study were collected from nine signalized intersections located in Amman, Irbid, and Zarqa, chosen due to their high population density and significant vehicular activity in Jordan. Each intersection conforms to a standard configuration with fixed cycle lengths, accommodating bidirectional pedestrian flow within three lanes per direction and a 60 km/h speed limit.
The primary crosswalk with the highest pedestrian traffic was selected from each intersection, totaling 2437 pedestrians observed and recorded on video during weekdays (Sunday to Thursday). This comprehensive dataset provided insights into various aspects of pedestrian crossing behavior, captured accurately through strategically positioned video cameras.
The video footage allowed for precise monitoring of pedestrian movements as they entered and exited the crosswalks. This approach ensured consistency in data collection across different intersections and minimized variability typically seen during peak pedestrian periods or weekends. The team analyzed the dataset meticulously to extract key variables encompassing pedestrian behaviors and characteristics relevant to the study's objectives.
They developed predictive models using the dataset employing four distinct ML techniques: ID3, ANN, RF, and SVM. These models were chosen for their ability to handle complex interactions within pedestrian data and effectively predict compliance with traffic regulations. Each ML technique utilized specific attributes from the video data to optimize accuracy and reliability in predicting pedestrian behavior at signalized intersections.
The ID3 algorithm facilitated classification by recursively partitioning data based on information gain, identifying significant attributes influencing pedestrian behavior. ANNs leveraged interconnected nodes to process and learn from pedestrian data patterns, which is crucial for predicting compliance and behavior. RFs, through ensemble learning, aggregated multiple ID3 to mitigate overfitting and enhance prediction accuracy. SVMs utilized kernel functions to classify pedestrian behaviors, effectively handling nonlinear relationships in the data.
Overall, these ML techniques contributed to a deeper understanding of pedestrian dynamics and safety at signalized intersections in urban Jordan. The insights gained from this study are instrumental for informing traffic management strategies and urban planning initiatives to improve pedestrian safety and compliance with traffic regulations.
Pedestrian Compliance Analysis
The study analyzed pedestrian behavior across nine signalized intersections in Jordan's urban centers of Amman, Irbid, and Zarqa, focusing on demographic and situational factors influencing compliance with traffic regulations. Results indicated a predominance of male pedestrians (61.7%) and highlighted varied spatial distributions of adults, children, and elders.
Most pedestrians crossed directly without pauses, often carrying personal items, with the majority (72.5%) avoiding vehicle interactions entirely. The dataset, comprising 2437 pedestrians observed on weekdays, facilitated detailed insights into pedestrian crossing behaviors through strategically positioned video cameras.
Classification models using ML techniques—ID3, ANN, RF, and SVM—were employed with a 10-fold cross-validation method to predict pedestrian compliance. Among these, RF achieved the highest accuracy (81.2%) and precision (80.9%), with a low false positive rate (FPR) of 25.2%. In contrast, SVM showed lower efficiency with a receiver operating characteristic (ROC) area under the curve of 50.4%.
The ID3 provided rule-based outputs identifying crucial variables influencing compliance, such as crossing type, demographics, and environmental conditions. These findings underscore the utility of ML in understanding and enhancing pedestrian safety at signalized intersections, offering actionable insights for urban planning and traffic management strategies in Jordanian cities.
Conclusion
To sum up, this study analyzed pedestrian behavior at urban crosswalks using four ML models: ANN, SVM, ID3, and RF. Data from 2437 pedestrians provided insights into pedestrian safety and compliance with traffic signals, revealing adults as predominant users near universities and shopping centers. Most pedestrians followed traffic laws by walking rather than running.
Random forest showed superior performance in accuracy, precision, and managing false-positive rates, offering valuable insights for urban planners and traffic authorities. Future research should focus on integrating these models into real-time traffic management systems and expanding data collection to enhance prediction accuracy in urban environments.
Metal Sulfide Deposit Classification Using Machine Learning