Public Safety Applications of Network Activity Data

通过收集与城市中不良事件有关的数据，可以促进主动公共安全，预测模型可以使用这些数据来识别和减轻导致伤害和死亡的模式。我们的目标是通过对匿名网络活动数据应用分析和机器学习（ML）来检测不良事件，从而创建低成本、有效的公共安全系统。当人们使用他们的应用程序时，他们的智能手机与蜂窝网络互动，这样他们就可以显示自己的位置。许多应用程序依赖于从该位置数据提取的用户移动信息，通常要求用户选择加入共享其位置数据。在这个项目中，我们感兴趣的是使用网络活动数据来保护个人隐私，并且不需要用户移动数据的模型。我们专注于三个用例中不良事件的ML预测和风险评估：Vision Zero试图将行人和骑自行车的死亡人数减少到零；消防和应急响应；还有911呼叫调度。该项目将调查和开发分析和机器学习算法，将网络活动数据与相关城市和其他数据结合起来处理，以预测和帮助应对随机时间和地点发生的不良事件。我们希望在一定的地理区域和时间间隔内预测这些事件，并通过类型、严重程度和其他属性来描述这些事件。因此，我们对预测行人和骑自行车者碰撞事件的发生和严重程度感兴趣。我们还对在不同天气条件下预测城市不同地区火灾的发生及其类型和影响感兴趣。对于911呼叫，我们的ML模型学习事件发生的模式及其类型和严重程度，通过自动选择并向调度员和第一响应者提供相关信息，帮助改善对紧急呼叫的响应。通过与加拿大几个城市的数据合作，我们的项目将通过改进规划、及时性、准确性和响应质量来帮助改善公共安全。

Proactive public safety can be facilitated by gathering data pertaining to adverse events in a city, and predictive models can use this data to recognize and mitigate patterns that lead to injuries and death. Our objective is to create low-cost, effective public safety systems by applying analytics and machine learning (ML) to anonymized network activity data to detect adverse events. As people use their apps, their smart phones interact with the cellular network and in doing so they indicate their location. Many applications depend on user mobility information extracted from this location data, typically requiring that a user opt-in to share their location data. In this project we are interested in models that use network activity data that preserves individual privacy, and that do not require user mobility data. We focus on ML prediction and risk assessment for adverse events in three use cases: Vision Zero which attempts to reduce pedestrian and bicyclist fatalities to zero; Fire and emergency response; and 911 call dispatching.The project will investigate and develop analytics and ML algorithms that process network activity data combined with relevant city and other data to predict and help respond to adverse events that occur at random times and places. We wish to predict these events over a range of geographic areas and time intervals, as well as to characterize these by type, severity, and other attributes. Thus, we are interested in predicting the occurrence and severity of pedestrian and cyclist crash events. We are also interested in predicting the occurrence of fires and their types and impacts in various districts of a city under various weather conditions. For 911 calls, our ML models learn patterns of event occurrences and their types and severity to help improve response to emergency calls by automatically selecting and providing relevant information to dispatchers and first responders. By working with data from several Canadian cities, our project will help to better public safety through improved planning, timeliness, accuracy, and quality of response.