Performance Analysis of Deep Neural Networks for Classification of Gene-Expression Microarrays

In recent years, researchers have increased their interest in deep learning for data mining and pattern recognition applications. This is mainly due to its high processing capability and good performance in feature selection, prediction and classification tasks. In general, deep learning algorithms have demonstrated their great potential in handling large scale data sets in image recognition and natural language processing applications, which are characterized by a very large number of samples coupled with a high dimensionality. In this work, we aim at analyzing the performance of deep neural networks for classification of gene-expression microarrays, in which the number of genes is of the order of thousands while the number of samples is typically less than a hundred. The experimental results show that in some of these challenging situations, the use of deep neural networks and traditional machine learning algorithms does not always lead to high performance results. This finding suggests that deep learning needs a very large number of both samples and features to achieve high performance.