Abstract:

: Many researchers have analyzed the high dimensional gene expression data for disease classification using several conventional and machine learning-based approaches, but still there exists some issues which make this task nontrivial. Due to the growing complexities of the unstructured data, the researchers focus on the deep learning approach, which is the latest form of machine learning algorithm. In the presented work, a kernel-based Fisher score (KFS) approach is implemented to extract the notable genes, and an improvised chaotic Jaya (CJaya) algorithm optimized convolutional neural network (CJaya-CNN) model is applied to classify high dimensional gene expression or microarray data. This model is tested on two binary class and two multi class standard microarray datasets. Here, the presented hybrid deep learning model (KFS based CJaya-CNN) has been compared with other standard machine learning classification models like CJaya hybridized multi-layer perceptron (CJaya-MLP), CJaya hybridized extreme learning machine (CJayaELM), and CJaya hybridized kernel extreme learning machine (CJaya-KELM). The suggested model is evaluated by classification accuracy percentage, number of significant genes selected, sensitivity and specificity values with receiver operating characteristic (ROC) curves. Eventually, the experimental outcomes obtained from the presented model has also been compared with the recent existing feature selection and classification models for a suitable research in analysing high dimensional microarray data. The presented model offered the classification accuracy percentage of 98.2, 99.96, 99.78, and 99.87 for colon cancer, leukemia, lymphoma-3, and small round blue cell tumor (SRBCT) datasets, respectively. All the experimental outcomes reveal that the KFS based CJaya-CNN model is outperforming. Hence, the presented method can be used as a dependable framework for disease classification.

Keywords: