Modelo preditivo de situações como apoio à consciência situacional e ao processo decisório em sistemas de resposta à emergência

• Autor(a) principal: Berti, Claudia Beatriz
• Data de Publicação: 2017
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/10119
• topic: Consciência da situação; Tomada de decisão; Resposta à emergência; Aprendizado de máquina; Classificação de texto; Situation awareness; Decision making; Emergency response; Machine learning; Text classification; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

Situation Awareness (SAW) is a concept widely used in areas that require critical decision making, and refers to the ability of an individual or team to perceive, understand and anticipate the future state of a current situation, which is influenced by the dynamicity and critical nature of events. SAW is considered as the main precursor of the decision-making process. In the emergency response area, obtaining and maintaining SAW requires a great effort from the human operator, the cognitive overload required in the activity, high level of stress involving the care, exhaustive shifts that may negatively reflect the care process and consequently the decision process as one all. Decision support systems that address aspects of the SAW can contribute to the enrichment and maintenance of the operator's SAW and in the decision-making process. Given this context, this work presents a Situational Predictive Model to systematize the development of modules to support the human operator's SAW in emergency response systems, which provides for the use of service models and protocols of institutions acting as prototypical situations. Objectively the model proposes the prediction and or the premature identification of the situation while the applicant has emergency assistance. A Conceptual Model was developed that guided the construction of the Predictive Model and will serve as basis for other developments. So-called human sensors and social sensors have become important sources of information especially in social networks. For the treatment of this data, text classifier methods are used with satisfactory results that cover the areas of education, security, entertainment, commercial, among others. For the emergency responses domain, object of this thesis, human sensors are the main source of information and machine learning techniques as text classifiers show important alternatives. In order to be validated, the Predictive Situations Model was implemented with the creation of a vocabulary based on the actual decision-making models of the Military Police of the State of São Paulo (PMESP) and the development of algorithms two classifying methods (Bag of Words and Naïve Bayes). Tests were performed with four different types of input instances (sentences). For all the metrics analyzed (accuracy, accuracy and coverage) the tests demonstrated superiority of the Naïve Bayes algorithm. The difference between the hit rates in relation to the Bag of Word algorithm for the class of instances with the highest degree of identification difficulty was over 37%. These results demonstrated good potential the Predictive Situations Model to collaborate with the existing systems of emergency services, allowing more attendance effectiveness and reduction of the cognitive overload that the attendants are routinely subjected to.