Análise e previsão de acidentes rodoviários usando data mining

• Autor(a) principal: Teixeira, Bruno Miguel Ferreira
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.22/14860
• topic: Road traffic crashes; Crashes prediction model; Machine learning; Artificial Neural Networks; Iowa; Acidentes rodoviários; Modelos de previsão de acidentes rodoviários; Redes neuronais

Road traffic crashes is an impactful problem in nowadays society, causing significant life and property losses. Due to the urbanization process across the world and the population’s growth, the number of crashes is also increasing. Predicting a crash severity and cost is an important step to better understand which causative variables have more influence and therefore, implement prevention measures that can reduce the number of crashes. Road traffic crashes predictions is a complex problem due to the high number of independent causative variables that contribute to the event. The used dataset contains crashes occurred in the State of Iowa in the recent years. Feature selection and data cleaning techniques are applied to improve the data quality and enhance the learning process. Previous research on the road safety field applied approaches that led to unsatisfactory results. Recent studies based on more complex approaches like neural networks had better results. This document’s work is based on deep learning, studying how the usage of deep neural networks can enhance previous results on road traffic crashes predictions taking causative variables as input. Various models are built using different optimization and activation functions. The evaluation is based on the comparison of these models.