Mining Network Hotspots with Holes: A Summary of Results

A Unified Framework for Robust and Efficient Hotspot Detection in Smart Cities

“…Thus, its higher recall does not come with stable control of false-positive rates, making the improvements less interesting. 7 In contrast, the unified framework is able to enforce the significance level and maintain a stable robustness against false positives. This is very important in urban and smart-city applications in which false positives often have a high cost (Section 1).…”

“…The spatial scan statistic by default detects hotspots that have a circular shape. In recent years, the shape family of hotspots has been extended to include many others, such as ellipses [20,21], rectangles [31,32], rings [7,8], and linear [13,35,39,41] or arbitrarily shaped hotspots [49]. These new hotspot shapes have mainly been developed to meet specific domain needs (e.g., ring-shaped hotspot detection for locating a serial criminal).…”

“…By default, it detects hotspots that have a circular shape. In recent years, most efforts in the data mining community have focused on extending this framework to detect hotspots of other geometric shapes (e.g., ring [7,8], linear [35,39], rectangle [31,32], ellipse [21]) to meet specific domain needs. Meanwhile, many other efforts have also been made to extend the original spatial scan statistic to detect multi-variate hotspots [22,28], spatiotemporal (i.e., emerging) hotspots [27,29], and continuous value (e.g., lifespan)-based hotspots [11,19].…”

“…Studies have also tried to analyze and mitigate the uncertainty (e.g., geographically aggregated data) and inaccuracy issues in hotspot detection [24,25]. Although various extensions exist, they either directly adopted the likelihood ratio framework provided by the spatial scan statistic (e.g., extensions for new shapes [7,8,21,32,39]) or enriched the likelihood ratio specifically for a certain application scenario (e.g., multi-variate-, temporal-, or continuous value-based hotspots [19,22,27,29]). In both cases, the methods do not attempt to change the fundamental way that the likelihood ratio is formulated or defined in the spatial scan statistic.…”

Transdisciplinary Foundations of Geospatial Data Science

“…However, these approaches do not guarantee completeness without grid partitioning (i.e., when points are treated as distinct 2D vectors). While all these methods focus on hotspots in Euclidean space, a few have been designed to model hotspots in network space (e.g., road network) [24,25]. These approaches currently only consider shortest paths between points during candidate enumeration rather than all simple paths due to scalability concerns.…”