SBIR Phase I: Predictive Analytics and Optimization for Crime Reduction and Police Service Improvement

SBIR第一阶段项目建议开发一个系统，该系统基于对过去犯罪的数据分析，预测未来可能犯罪的类型、地点和时间，并自动建议适当的警察部队部署。建议的研究在两个基本方面是创新的：首先，该系统将犯罪事件和警察部队部署视为一个整体问题，并创造犯罪预测和适当的资源分配，以达到最高效率。其次，鲁棒优化的最新进展将用于处理资源分配过程中的任何不确定性，并产生在偏离预测或分配计划的情况下产生良好结果的部署计划。该系统将使用最先进的预测分析和鲁棒优化技术相结合或相互嵌入，以实现更高程度的预防犯罪效率，并在1至30天的时间范围内设计最佳单元部署。该方法是对现有系统的重大改进，现有系统使用预测分析预测犯罪，然后将结果传递给不同的调度系统或人工来设计单元部署，而不考虑不确定性或预测误差。这项计划的更广泛影响是，在可接受的确定程度下，确定一系列可能发生特定类型罪行的地点和时间，并建议部署适当的部队，从而降低警务处的罪案指数。大多数警察没有足够的资源部署在每一个可能的犯罪附近；他们需要工具来帮助他们通过对犯罪指数总体上影响最大的犯罪做出反应来选择单位部署。如今市场上很少有这样的工具；他们中的大多数专注于创造每周或每月的预测，或者追踪已知的罪犯。一些更先进的系统需要昂贵的硬件和训练有素的人员来操作，这使它们超出了规模较小的警察局的能力。拟议的系统将是高度可定制的，能够与几乎任何类型的数据存储库连接，同时通过现场或托管访问负担得起。因此，它将提高预防犯罪率，缩短反应时间，提高预算有限的警察部门的效率。从长远来看，它还将使战略单位部署和有效的人员规划成为可能。

This SBIR Phase I project proposes to develop a system that, based on data analysis of past crimes, forecasts the type, location and timing of probable future crimes and automatically suggests appropriate police unit deployment. The proposed research is innovative in two fundamental ways: First, the system will view crime occurrences and police unit deployment as a holistic problem and create both a crime forecast and the appropriate resource allocation for maximum efficiency. Second, recent advances in Robust Optimization will be used to handle any uncertainties during the resource allocation process and produce a deployment plan that yields good results under scenarios that deviate from the forecast or the allocation plan. The system will use state-of-the-art Predictive Analytics and Robust Optimization in combination or embedded within each other in order to achieve a much higher degree of effectiveness in crime prevention and to design optimal unit deployment for time horizons between one and thirty days. The proposed approach is a substantial improvement over currently available systems, which forecast crime using predictive analytics and then pass the results to either a different scheduling system or a human to design the unit deployment without any consideration for uncertainty or forecast error. The broader impact of this project is to lower the crime index of police departments (PDs) by determining a series of locations and times where crimes of specific types are probable, with an acceptable degree of certainty, and suggesting deployment of appropriate units. Most PDs do not have enough resources to deploy in the vicinity of every probable crime; they need tools to help them select unit deployment by responding to the set of crimes that will have the most effect on the crime index on aggregate. They are few such tools available in the market today; most of them concentrate on creating weekly or monthly predictions, or on tracking known criminals. Some of the more advanced systems require expensive hardware and well-trained personnel to operate, putting them beyond the reach of smaller Police Departments. The proposed system will be highly customizable and capable of connecting with almost any type of data repository, while being affordable through on site or hosted access. Therefore, it will improve crime prevention rates, decrease response time and increase the effectiveness of budget-constrained Police Departments. In the long term it will also enable strategic unit deployment, and efficient personnel planning.