MicroarrayCancerNet: Hybrid optimized deep learning with integration of graph CNN with 1D-CNN for cancer classification framework using microarray and seq expression data

The key difficulty lies in accurately classifying the relevant genes through analysis and selection. A variety of
methods are used to classify the genes. However, in the selection of numerous genes in the huge dimensional
microarray data, only a limited amount of success has been achieved. Thus, this study focuses on designing a new
cancer classification framework. In the initial stage, the microarray and seq expression information is attained
from the standard datasets. Next, the pre-processing is performed using NAN removal and the missing value
removal from the samples to convert it into a numeric feature matrix for making the data suitable for further
levels of processing. Then, the Modified Sandpiper Optimization Algorithm (MSOA) is suggested for confirming
the optimal gene from the pre-processed information. Finally, the chosen optimal gene is fed to the cancer
classification stage, where the Hybrid Deep Learning Framework (HDLF) is suggested by incorporating the Graph
Convolutional Neural Network (GCNN) with One-Dimensional Convolutional Neural Networks (1D-CNN). The
parameters of both Graph CNN and 1D-CNN are tuned via the same MSOA. Finally, the experimental results
confirm that the developed model performs well compared to existing machine learning and currently utilized
deep learning methods for cancer classification. The precision of the proposed model is 91.78 %.