Machine Learning, Structured Data, Unstructured Data, Public Safety, Big Data
Public safety should be strategically designed to minimize crime and ensure a higher level of security. It is found that crime data analysis contributes to the establishment of actions that should be taken to ensure this higher level of security. In these analyses, structured data from government agencies and unstructured data from social networks such as Twitter are considered, but not both together. It is found that there is no incorporation and integration of this data, which would allow us to increase in the accuracy of data analysis methods such as machine learning algorithms. Thus, this paper proposes a hybrid data analysis methodology, considering the integration of structured data and unstructured data, which would allow data analysis algorithms, such as machine learning, for example, to perform the necessary analysis for public safety. The integration happens in two main spheres, the first from the absorption and analysis of structured data made available by government agencies, and the second from the absorption, classification, and analysis of unstructured data, coming from digital platforms, as is the case of Twitter, the Where Was I Robbed platform and CityCop. With this, it becomes possible to transform and incorporate this data into a single repository base. Based on this hybrid methodology, a decision support system (DS.Security) was built so that analysis can occur in public security. DS.Security has three models to illustrate the applicability of this methodology: the first is for neighborhood classification; the second is for criminal profile; and the third is for neighborhood classification, taking violence against women into consideration. As a result, the accuracy rate of machine learning algorithms such as J.48, ExtraTrees, Naive Bayes, and Support Vector Machine increased by about 10% compared to applying these algorithms to only structured data in the models associated with the decision support system.