Redução da dimensionalidade em dados da saúde por meio de combinação de algoritmos de seleção de atributos

• Autor(a) principal: Villar, Gustavo Henrique de Oliveira
• Data de Publicação: 2022
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/216220
• topic: Data mining; Knowledge discovery; Health data; Data prediction; Big Data; Feature selection; Classification algorithms; Machine learning; Banco de Dados; Mineração de dados; Extração de conhecimento; Saúde; Predição de dados; Seleção de atributos; Algoritmos de classificação; KNN; J48

Technological advances in the last years allowed the creation of tools capable of collecting a significant amount of data in the different sectors of society, among those sectors is the health industry. This abundance of data has a vast potential of generating knowledge after it has been processed; however, the volume of this data makes it unfeasible to manually explore all this capacity. In this scenario it is necessary the application of well-defined automation techniques that are able to extract knowledge from the data. One of these techniques is the process of Data Mining as a way to predict diagnosis using element classification on a dataset, methodology that has been reaching uplifting results given that it makes for a more accurate and faster diagnostic process when compared to those that do not rely on computer assisted decision making. Nevertheless, the characteristics of health data, such as, elevated volume and high dimensionality create challenges on the process of knowledge discovery making it less efficient and with a higher computational cost, as a way to solve this problem it is possible to apply feature selection algorithms that reduce the number of present attributes and allow us to have a better understanding of how much a single attribute can affect the final result. Therefore, the objective of this work is to create a scientific contribution based on the application of several feature selection methods associated to a couple of classification algorithms on multiple datasets that explore both breast cancer and cardiac diseases. The results show that feature selection techniques can significantly increase the accuracy of the classification of health data with very little or no losses on execution time, given that in most cases it’s actually doable do reduce the execution time, leading to faster and more accurate predictions a very important aspect when taking the medical field in consideration.