An approach for cancer classification using optimization driven deep learning

The normal and cancer cell tissues exhibit different gene expressions. Therefore, gene expression data are the effective source for cancer classification, in which the usage of the original gene expression data is challenging due to their high dimension and small size of the data samples. This article proposes a fractional biogeography-based optimization-based deep convolutional neural network (FBBO-based deep CNN) for cancer classification. The developed FBBO is the integration of the fractional calculus (FC) in the biogeography-based optimization (BBO), which aims at determining the optimal weights for tuning the deep CNN. Initially, the gene expression data is pre-processed and subjected to dimensional reduction using the probabilistic principal component analysis (PPCA). The selected features are used for cancer classification enabling a high degree of robustness and accuracy. The experimental analysis using the Colon dataset and Leukemia dataset reveals that the proposed classifier acquired maximal accuracy, sensitivity, specificity, precision, and F-Measure of 0.98.