Bangladesh Road Traffic Accident Dataset (2007–2024): Multi-Source Integration.

This dataset presents a comprehensive compilation of 49,566 road traffic accident records from Bangladesh, meticulously consolidated from four authoritative institutional sources: the Accident Research Institute (ARI) of BUET, the Bangladesh Road Transport Authority (BRTA), the Dhaka Metropolitan Police (DMP), and the Military Police (MP) Unit, along with a primary dataset collected directly through structured reporting forms. Spanning a 17-year period (2007–2024), it integrates diverse data streams into a unified, structured, and research-ready resource for traffic safety and accident analysis. Specifically, ARI contributed historical accident records from 2007–2021, BRTA provided official national accident statistics for 2024, DMP supplied detailed urban traffic incident logs from 2020–2024, and the MP Unit furnished verified accident reports covering the same recent period. Additionally, the primary dataset collected via forms captures incident details directly from field reports, ensuring high granularity and accuracy in recorded attributes.The dataset captures a rich variety of attributes, including:Temporal Information: Exact date, time, and year of each recorded accident.Geospatial Data: Detailed location descriptions with latitude and longitude coordinates for spatial analysis.Incident Narratives: Text-based summaries describing the circumstances of each accident.Vehicle and Victim Data: Vehicle types, collision categories, number of vehicles involved, casualty counts, injury details, and victim classifications.Environmental Conditions: Light levels, weather conditions, and road surface status at the time of the incident.Road and Infrastructure Attributes: Road classification, nature, connection type, surface condition, junction types, and traffic control measures.Accident Severity Indicators: Death and injury counts, as well as accident intensity classifications. Preliminary statistical and correlation analyses conducted on the dataset reveal generally weak linear relationships between predictive features and accident severity, suggesting that traditional linear modeling techniques may have limited predictive capability. Consequently, advanced ensemble-based machine learning approaches, which can capture non-linear interactions, mitigate overfitting, and improve predictive robustness, are recommended for modeling and forecasting purposes.This dataset offers a robust empirical foundation for diverse research and policy applications, including traffic safety assessments, accident hotspot mapping, predictive risk modeling, urban transport infrastructure planning, and evidence-based road safety policy formulation. The inclusion of both institutional records and primary field-collected data ensures a level of depth, reliability, and coverage that makes this dataset an invaluable resource for academics, policymakers, urban planners, and data scientists addressing road safety challenges in Bangladesh.

本数据集收录了孟加拉国境内共计49566条道路交通事故记录，经严谨整合自四大权威机构来源：孟加拉国工程技术大学（Bangladesh University of Engineering and Technology，BUET）事故研究所（Accident Research Institute, ARI）、孟加拉国道路运输局（Bangladesh Road Transport Authority, BRTA）、达卡大都会警察局（Dhaka Metropolitan Police, DMP）及宪兵分队（Military Police, MP），同时包含通过结构化报告表单直接采集的原始数据集。
该数据集覆盖2007至2024年共17年的时间跨度，将多源数据整合为统一、结构化且适用于交通安全与事故分析的研究就绪型资源。

具体而言，事故研究所（ARI）提供了2007年至2021年的历史事故记录，孟加拉国道路运输局（BRTA）提供了2024年的官方全国事故统计数据，达卡大都会警察局（DMP）提供了2020年至2024年的详细城市交通事件日志，宪兵分队（MP）则提供了同期的经核实事故报告。此外，通过表单采集的原始数据集直接从现场报告中提取事件细节，确保了记录属性的高细粒度与准确性。

本数据集包含丰富多样的属性字段，具体包括：
- 时间信息：每起记录事故的精确日期、时刻与年份
- 地理空间数据：用于空间分析的详细位置描述及经纬度坐标
- 事件叙述：描述每起事故发生场景的文本摘要
- 车辆与受害者数据：车辆类型、碰撞类别、涉事车辆数量、伤亡人数、伤情细节及受害者分类
- 环境条件：事故发生时的光照条件、天气状况与路面状态
- 道路与基础设施属性：道路等级、道路性质、连接类型、路面状况、路口类型及交通管控措施
- 事故严重程度指标：死亡与受伤人数，以及事故强度分级

针对本数据集开展的初步统计与相关性分析显示，预测特征与事故严重程度之间普遍存在较弱的线性关联，这表明传统线性建模技术的预测能力较为有限。因此，建议采用可捕捉非线性交互关系、缓解过拟合并提升预测鲁棒性的先进集成机器学习方法，用于事故建模与预测任务。

本数据集为多元研究与政策应用提供了坚实的实证基础，涵盖交通安全评估、事故热点制图、预测风险建模、城市交通基础设施规划以及基于证据的道路安全政策制定等场景。纳入官方机构记录与原始实地采集数据的双重数据来源，确保了数据集的深度、可靠性与覆盖范围，使其成为孟加拉国道路安全问题研究者、政策制定者、城市规划者与数据科学家的宝贵资源。