An Agentic Approach for Active Data Collection, Travel Behavior Modeling, and Weather-Sensitive Demand Prediction
A study proposes a three-agent workflow integrating conversational data collection, structured data processing, and behavioral prediction. A chatbot-administered, image-augmented stated-preference survey collected mode choices from student commuters across five predefined weather scenarios, yielding 454 respondent-scenario observations. Nine locally deployed large language models (LLMs), ranging from 2 to 35 billion parameters, were evaluated across four zero-shot prompt-and-context conditions and extended through persona, few-shot, and vision-based configurations. Random forest achieved 69.6% five-class accuracy, while the best text-only zero-shot LLM reached 69.9% without task-specific fitting.
Travel behavior research increasingly combines digital data collection with predictive modeling, yet these stages are often developed and evaluated separately. This study proposes a three-agent workflow integrating conversational data collection, structured data processing, and behavioral prediction. A chatbot-administered, image-augmented stated-preference survey collected mode choices from student commuters across five predefined weather scenarios, yielding 454 respondent-scenario observations. Weather-related associations were analyzed using a multinomial logit model, while logistic regression and random forest provided machine-learning benchmarks. Nine locally deployed large language models (LLMs), ranging from 2 to 35 billion parameters, were evaluated across four zero-shot prompt-and-context conditions and extended through persona, few-shot, and vision-based configurations. Random forest achieved 69.6% five-class accuracy, while the best text-only zero-shot LLM reached 69.9% without task-specific fitting. Habitual travel information produced the most consistent gains, Expert framing generally outperformed Role-Play, and persona information was most useful when habitual travel information was absent.
The study demonstrates that zero-shot LLMs can match or slightly exceed a tuned random forest on a five-class travel mode prediction task, suggesting that LLMs can serve as competitive baselines for behavioral prediction without task-specific training. The three-agent workflow separates data collection, processing, and prediction, which may improve modularity and reproducibility. The use of locally deployed models (2-35B parameters) indicates feasibility of privacy-preserving, on-device inference for travel behavior modeling.
This research highlights a growing trend toward agentic AI systems that combine conversational interfaces with predictive analytics for transportation and urban planning. The ability to collect stated-preference data via chatbots and immediately apply LLM-based prediction could streamline survey-based demand modeling. The focus on local deployment suggests interest in data privacy and reduced cloud dependency for public-sector and mobility applications.
The approach could reduce the cost and time of travel behavior surveys by automating data collection and analysis. For transportation agencies and mobility companies, an LLM-based prediction pipeline may offer a flexible alternative to traditional statistical models, especially when data is limited or privacy constraints favor local processing. The open-source nature of the evaluated models suggests low adoption barriers.
Next signals to watch include whether the proposed agentic workflow is adopted in real-world travel demand models, whether larger or multimodal LLMs further improve accuracy, and whether similar agentic data collection methods are applied to other domains such as energy demand or public health. Replication studies with broader populations and real-world validation would strengthen the findings.