Trust Estimation of Real-Time Social Harm Events

Abstract

Social harm involves incidents resulting in physical, financial, and emotional hardships such as crime, drug overdoses and abuses, traffic accidents, and suicides. These incidents require various law-enforcement and emergency responding agencies to coordinate together for mitigating their impact on the society. With the advent of advanced networking and computing technologies together with data analytics, law-enforcement agencies and people in the community can work together to proactively reduce social harm. With the aim of effectively mitigating social harm events in communities, this thesis introduces a distributed web application, Community Data Analytic for Social Harm (CDASH). CDASH helps in collecting social harm data from heterogenous sources, analyzing the data for predicting social harm risks in the form of geographic hotspots and conveying the risks to law-enforcement agencies. Since various stakeholders including the police, community organizations and citizens can interact with CDASH, a need for a trust framework arises, to avoid fraudulent or mislabeled incidents from misleading CDASH. The enhanced system, called Trusted-CDASH (T-CDASH), superimposes a trust estimation framework on top of CDASH. This thesis discusses the importance and necessity of associating a degree of trust with each social harm incident reported to T-CDASH. It also describes the trust framework with different trust models that can be incorporated for assigning trust while examining their impact on prediction accuracy of future social harm events. The trust models are empirically validated by running simulations on historical social harm data of Indianapolis metro area.