Scholarlink

Disappearances in Peru remain underexamined: over 50% of reported individuals are never located, and institutional gaps persist. We present a two-stage multimodal framework using RENIPED structured records, narrative descriptions, and facial images. First, Gaussian Mixture Models uncover five behavioral profiles; clustering performs best with structured attributes, revealing patterns tied to reporting delays, ethnicity, and regional origin. Second, a multimodal regressor—using only structured and image data—estimates the likelihood of voluntary vs. involuntary disappearances, achieving a mean absolute percentage error of 1.12% without relying on text. Findings show that machine learning can surface latent patterns and support investigations in data-limited settings. Outcomes are shaped by gender, geography, and reporting lags, highlighting the need for more responsive, context-sensitive interventions. Code and implementation: https://github.com/yoce3/MissingPersonAnalisys .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Uncovering Disappearance Dynamics in Peru: A Two-Stage Multimodal Framework for Behavioral Segmentation and Voluntariness Prediction in Missing-Person Reports

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alejandro Aybar-Flores,
  • Rocío Maehara,
  • Luis Benites,
  • Miguel Muñoz,
  • Jose Carlos Salinas

摘要

Disappearances in Peru remain underexamined: over 50% of reported individuals are never located, and institutional gaps persist. We present a two-stage multimodal framework using RENIPED structured records, narrative descriptions, and facial images. First, Gaussian Mixture Models uncover five behavioral profiles; clustering performs best with structured attributes, revealing patterns tied to reporting delays, ethnicity, and regional origin. Second, a multimodal regressor—using only structured and image data—estimates the likelihood of voluntary vs. involuntary disappearances, achieving a mean absolute percentage error of 1.12% without relying on text. Findings show that machine learning can surface latent patterns and support investigations in data-limited settings. Outcomes are shaped by gender, geography, and reporting lags, highlighting the need for more responsive, context-sensitive interventions. Code and implementation: https://github.com/yoce3/MissingPersonAnalisys .