Redução de dimensionalidade em base de dados de microarranjo usando rede neural autocodificadora

Abstract

Machine Learning Algorithms have been increasingly used by its ability to learn from large volumes of data, such as gene expression data obtained by the microarray technique. A characteristic of the microarray data is that, generally, it is formed by a large amount of attributes and a few samples. It is known that data with high dimensionality can have redundant and often irrelevant attributes, it can hinder the learning process and the performance of predictions. Methods of dimensionality reduction are used to reduce the amount of attributes of the database. Autoencoder Neural Networks can be adapted and used for attribute extraction and, consequently, dimensionality reduction. This job aims to use an autoencoder neural network combined with a classifying neural network to perform a dimensionality reduction. For this, experiments will be carried out in five databases. The results were evaluated through the hit rate of the classifiers KNN, SVM, Naive Bayes and Decision Tree. The results show that the method created very representative smaller databases.