Scholarlink

<p>Human-bear conflicts involving Asiatic black bears (Ursus thibetanus) and brown bears (Ursus arctos) are a growing global concern due to factors, such as food scarcity, urbanization, and environmental changes. In 2023, Japan’s Akita Prefecture saw a surge in bear encounters from an annual average of 800 to 3,910, resulting in severe injuries and underscoring the urgent need for preventative measures. This study proposes a machine learning approach to predict bear encounters, utilizing temporal features (e.g., past bear encounters), environmental features (e.g., land cover type, weather conditions, road conditions, elevation, acorn abundance index), and demographic features (e.g., population distribution). We constructed a fine-grained dataset by integrating bear encounter data with geospatial, statistical, and meteorological sources, structured on a 1&#xa0;km <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41060_2025_866_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 1&#xa0;km grid with daily resolution for detailed training and validation. The model was trained on data from 2021 and 2022 and tested on 2023 encounter data. ExtraTrees achieved the highest performance in Accuracy (0.637), Precision (0.635), Recall (0.636), and Prec@C1 (0.595), significantly outperforming the baseline (kNN). An ablation study confirmed the additive contribution of each feature group, especially land cover, population, and encounter history. Furthermore, a leave-one-location-out evaluation demonstrated that the model maintained stable performance across unseen municipalities, supporting its spatial generalizability. SHAP analysis revealed that past encounters, land cover types (e.g., built-up areas, bamboo forests), sparse and elderly populations, and lower elevations are key predictors.</p>

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

Bear warning: predicting encounters using temporal, environmental, and demographic features

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Shin Nakamoto,
  • Yusuke Fukazawa

摘要

Human-bear conflicts involving Asiatic black bears (Ursus thibetanus) and brown bears (Ursus arctos) are a growing global concern due to factors, such as food scarcity, urbanization, and environmental changes. In 2023, Japan’s Akita Prefecture saw a surge in bear encounters from an annual average of 800 to 3,910, resulting in severe injuries and underscoring the urgent need for preventative measures. This study proposes a machine learning approach to predict bear encounters, utilizing temporal features (e.g., past bear encounters), environmental features (e.g., land cover type, weather conditions, road conditions, elevation, acorn abundance index), and demographic features (e.g., population distribution). We constructed a fine-grained dataset by integrating bear encounter data with geospatial, statistical, and meteorological sources, structured on a 1 km \(\times \) × 1 km grid with daily resolution for detailed training and validation. The model was trained on data from 2021 and 2022 and tested on 2023 encounter data. ExtraTrees achieved the highest performance in Accuracy (0.637), Precision (0.635), Recall (0.636), and Prec@C1 (0.595), significantly outperforming the baseline (kNN). An ablation study confirmed the additive contribution of each feature group, especially land cover, population, and encounter history. Furthermore, a leave-one-location-out evaluation demonstrated that the model maintained stable performance across unseen municipalities, supporting its spatial generalizability. SHAP analysis revealed that past encounters, land cover types (e.g., built-up areas, bamboo forests), sparse and elderly populations, and lower elevations are key predictors.