Towards Efficient and Evidence-grounded Mobility Prediction with LLM-Driven Agent

2026-06-03

AI Summary

Individual-level mobility prediction is often hindered in scenarios with weak or conflicting signals due to insufficient evidence utilization and opaque decision-making. This work proposes a training-free, LLM-driven agent framework that formulates next-location prediction as an adaptive, evidence-guided decision process: it employs a fast path for routine behaviors and dynamically invokes multi-source evidence—including trajectory history, behavioral regularities, stay-move probabilities, and geographic context—for iterative reasoning in ambiguous cases. To the best of our knowledge, this is the first approach to achieve multidimensional evidence fusion within an LLM agent for mobility prediction. The method attains state-of-the-art performance among training-free LLM-based approaches across three datasets, with GPT-4 achieving 71.42% Acc@1 on the BW dataset and improving accuracy by nearly 20 percentage points over statistical baselines in non-fast-path cases.

Abstract

Individual-level mobility prediction is central to urban simulation, transportation planning, and policy analysis. Supervised sequence models achieve strong accuracy but require task-specific training and offer limited decision-level transparency. Recent LLM-based methods improve interpretability, yet mostly rely on static prompts and single-pass inference, limiting their ability to seek additional evidence when mobility signals are weak or conflicting. We propose \method{}, a training-free LLM-driven agent framework that formulates next-location prediction as adaptive evidence-controlled decision making. \method{} resolves routine cases through a fast path based on historical regularity, while ambiguous cases trigger iterative tool use over recent trajectories, historical behavior, stay-move likelihood, and geographical evidence. Across three mobility datasets, AgentMob achieves the strongest overall performance among training-free LLM-based methods, with GPT-5.4 reaching 71.42\% Acc@1 on BW, 33.14\% on YJMob100K, and 33.50\% on Shanghai ISP. On BW non-fast-path cases, the LLM controller improves Acc@1 from 30.65\% to 48.62\% over a same-tool statistical baseline, showing that its main benefit lies in resolving ambiguous predictions through adaptive evidence gathering. Our code is available at https://github.com/Unknown-zoo/AgentMob.

Problem

• mobility prediction
• LLM-driven agent
• evidence gathering
• decision transparency
• ambiguous prediction

Innovation

• LLM-driven agent
• evidence-grounded reasoning
• adaptive decision-making
• mobility prediction
• training-free framework