Abstract: With the accelerating aging process in China, leveraging social forces to safeguard the safety of missing elderly has become a crucial measure in building an elderly-friendly society. The instability of volunteer mobilization efforts and the extreme time-sensitivity of locating these vulnerable individuals pose a dual challenge that demands innovative solutions. However, the current lag in digital and intelligent intervention measures within social volunteer organizations has led to issues such as inadequate prediction accuracy for time consumption during rescue operations and delayed responses to police alerts. These challenges hinder further improvements in the effectiveness of social rescue systems. Establishing robust, data-driven volunteer rescue capabilities is therefore paramount for effectively addressing these urgent cases in the future. This study tackles these intertwined issues by constructing a data-driven search and rescue system for social volunteer organizations.
This study establishes a data-driven search and rescue system for missing elderly persons in social volunteer organizations through investigations of WZ Emergency Rescue Team and historical data mining. Machine learning algorithms are introduced to conduct search duration prediction based on real case characteristics. Subsequently, a system optimization model is developed with the objective of minimizing full-chain circulation time of police alert information, accompanied by a solution algorithm tailored to model characteristics. The system integrates machine learning and the Ant Colony Algorithm, aiming to achieve highly accurate predictions of the time required for such rescues and significantly accelerate the flow of critical information throughout the entire emergency response chain.
This paper makes three key academic contributions to addressing missing seniors in China’s aging society: first, it identifies critical factors like dementia severity, advanced age, and reporting time that significantly influence volunteer rescue decisions through real-world case data analysis; second, it develops a scalable, time-sensitive machine learning model to predict search duration, enhancing the precision of volunteer mobilization alerts; and third, it creates an algorithm that optimizes multi-stage emergency information flow, providing volunteer organizations with a practical, data-driven framework for improving dementia-related rescue operations.
To achieve precise time-consumption predictions for missing senior rescues, this research rigorously evaluated leading algorithms (Decision Tree, Random Forest, Gradient Boosting, XGBoost), finding Random Forest superior in accuracy. Enriching historical rescue records further enhanced its prediction accuracy and stability. Integrating these predictions into mobilization guidance helps volunteers better balance professional and volunteer commitments, boosting rescue call credibility and encouraging more stable participation. The adaptable prediction module holds significant potential for broader application in assessing rescue workloads and expanding AI use in volunteer emergency response. Building on this predictive foundation, the study developed a data-driven search and rescue system for missing elderly persons in social volunteer organizations optimizing the dispatch of resources and critical case information. Applying the Ant Colony Algorithm demonstrably improved efficiency, reducing average completion time by 4.72%, standard deviation by 2.70%, and coefficient of variation by 4.37%. This faster information flow enables quicker mobilization for time-critical rescues over wider areas. This optimization module is a significant step towards leveraging technology within the collaborative “government-led, society-participated” emergency response mode, with scalability promise for larger or more complex response chains.
This research provides theoretical and methodological references for technical interventions in social rescue operations, offering significant practical application value for other types of social emergency response scenarios. Future research directions can consider the impact of standardizing case record protocols within social volunteer organizations on the robustness and applicability of predictive models. An extended model incorporating the evolving impact of widespread smart anti-wandering devices for seniors, which may enable shorter response networks and alter case information flow dynamics, can also be considered. These need to be further studied and demonstrated to establish highly efficient “early rescue, swift resolution” mode.

Key words: missing elderly, emergency management, social rescue, machine learning

摘要： 随着我国老龄化进程加快,发挥社会力量维护失踪老人生命安全是构建老人友好型社会的重要举措。但是,现阶段社会志愿组织数智化干预措施发展滞后,导致搜救时耗预测精度不足与警情响应迟缓等问题,阻碍了社会搜救系统效能进一步提升。为此,本研究基于对WZ应急救援队调研与历史数据挖掘,建立数据驱动的社会志愿组织老人走失搜救系统,引入机器学习算法开展基于真实案情特征的搜救时耗预测,在此基础上以警情信息全链条流转时间最小化为目标构建了系统优化模型,并结合模型特征给出求解算法。结果表明,Random Forest模型的搜救时耗预测效果最为出色,具备较高精度与稳定性;结合蚁群算法的警情信息调度优化模型可以实现警情在多救援环节多信息处理单元之间的流转速度优化,且优化效果随警情复杂性提升而提升。本研究为社会救援力量采用技术性干预措施提供了理论方法参考,对于其他类型社会救援事件具有较高的实践推广价值。

关键词: 老人走失, 应急管理, 社会救援, 机器学习