Using Deep Learning to Classify Class Imbalanced Gene-Expression Microarrays Datasets.

Performance of deep learning neural networks to classify class imbalanced gene-expression microarrays datasets is studied in this work. The low number of samples and high dimensionality of this type of datasets represent a challenging situation. Three sampling methods which have shown favorable results to deal with the class imbalance problem were used, namely: Random Over-Sampling (ROS), Random Under-Sampling (RUS) and Synthetic Minority Oversampling Technique (SMOTE). Moreover, artificial noise and greater class imbalance were included in the datasets in order to analyze these situations in the context of classification of gene-expression microarrays datasets. Results show that the noise or separability of the dataset is more determinant than its dimensionality in the classifier performance.