Bottleneck Feature Extraction for Gene Expression Using Deep Learning

Abstract

Cancer is one of the deadly diseases that is touching the masses. Gene expression data consists of fewer samples and more features, which makes it difficult to handle. So, for this purpose, various dimensionality reduction techniques are available in literature. With the help of these methods, important features are extracted from the data and then later cancer classification is done. The suggested approach involves combining pre-trained models VGG16 and VGG19. VGG19 is positioned between two VGG16 models. Once the relevant features have been extracted from the data, XGBoost (extreme gradient boosting) is employed as a classifier to categorise the data into five cancer classifications. The suggested technique has been compared to current methods such as VGG16, VGG19, ResNet50, and Inception V3. It has been observed that the proposed method exhibits lower mean squared error (MSE) and higher accuracy compared to the other methods.