Avaliação de predição de violência contra a mulher através de estratégias de aprendizado de máquinas

• Ano de defesa: 2023
• Autor(a) principal: Costa, Ana Carolina Nepomuceno
• Orientador(a): Cavalcante, Charles Casimiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Algoritmos supervisionados; Aprendizado do computador; Violência contra as mulheres; Supervised algorithms; Machine learning; Violence against women; Feature selection
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/74007

Violence against women is a educational, public health and safety problem that affects one in three women worldwide, according to the World Health Organization (WHO). In addition, it is clear that violence against women begins prematurely, as one in four women between 15 and 24 years have already reported violence committed by their partner when they were in a relationship. In addition, 1.48 million women reported violence between 2010 and 2018, according to the Igarapé Institute. In view of these worrying numbers, this work aims to predict violence against women and find patterns of violence in order to reduce these alarming rates and assist in results measurement. For this, models were proposed to obtain metrics for classifying and identifying characteristics that result in violence. Therefore, this research is focused on identifying which attributes are most important for a good result of the model, using Feature Selection and recognizing patterns of violence through exploratory analysis of data obtained by the Graduate Program in Economics (CAEN/ UFC), in which the questionnaires are from the Survey of Socioeconomic Conditions and Domestic and Family Violence against Women (PCSVDF-Mulher, in Portuguese).The results indicate high performance, around 81%, on the classification of violence events, showing a real possibility to contribute with predictive models to formularies of violence information.