Comparative Analysis of Dimensionality Reduction Techniques Applied to Disease Classification Tasks

Dimensionality reduction techniques are useful in Machine Learning due they can allow to improve classifiers performance or even use low complexity models by helping to deal with redundancy and noise issues which may be present in the problems datasets. In this work there are used Principal Component Analysis, AutoEncoder and Genetic-Algorithm-based dimensionality reduction methods to decrease the feature number of several diseases datasets for medical diagnosis. The different reduced datasets are used for training and testing K-Nearest Neighbours-based classifiers to statistically analyze in order to find out which method gives better accuracy results and verify if it preserves or improves the performance than the same kind of classifier which is trained and tested with the whole dataset.