Automated density-based clustering of spatial urban data for interactive data exploration

Data collected in this repository is structured as follows

code/: The folder containing "RiskMap" web app project (maven). In particular, the code for automatic parameter estimation can be found in code/RiskMap/src/thesis/servlet folder.

Automatic parameter estimation of DBSCAN can be found in DBSCANServlet.java.

Automatic parameter estimation of HDBSCAN can be found in HDBSCANServlet.java.

paper/: Formal description of the algorithm and evaluation result.

presentation/: PDF of paper presentation in certain conference or venue.

The data collected in this repository relates to the paper "Automated density-based clustering of spatial urban data for interactive data exploration". This paper presents a method to automatically estimate parameters for density-based clustering based on data distribution. It also includes several techniques for visualizing the clusters over a map, useful for interactive data exploration. The proposed method enables parameter estimation to automatically adapt to multiple resolutions, allowing the clusters to be recomputed and visualized interactively at query time with the changes of zoom levels and panning of the map. We apply a voting scheme with existing cluster indices to rank the clustering results. The framework of multi-resolution density-based clustering and visualization is implemented and evaluated using a real-world road crash datasets.

本仓库所收录的数据组织结构如下：
code/：存放"RiskMap"Web应用项目（基于Maven构建）的文件夹。具体而言，自动参数估计的相关代码位于code/RiskMap/src/thesis/servlet目录下。其中，DBSCAN（密度聚类算法）的自动参数估计代码存放在DBSCANServlet.java中，HDBSCAN（层次化密度聚类算法）的自动参数估计代码存放在HDBSCANServlet.java中。
paper/：存放算法的正式学术说明与聚类评估结果。
presentation/：用于特定学术会议或场合的论文演示PDF文档。
本仓库所收录的数据关联论文《面向交互式数据探索的空间城市数据自动化密度聚类（Automated density-based clustering of spatial urban data for interactive data exploration）》。该论文提出了一种基于数据分布自动估计密度聚类（Density-based Clustering）参数的方法，同时集成了多种用于在地图上可视化聚类结果的技术，可有效支持交互式数据探索任务。所提方法可实现参数估计的自动多分辨率适配，支持在地图缩放、平移时，于查询时刻自动重新计算并交互式可视化聚类结果。研究采用结合现有聚类指标的投票方案对聚类结果进行排序。本研究基于真实世界的道路交通事故数据集，实现并评估了多分辨率密度聚类与可视化的完整框架。