Nowadays, high dimension microarray data are largely used in the field of biological research, critical disease prediction, and diagnosis. The microarray data contain a large number of highly correlated gene expressions in which feature extraction is required for accurate identification of informative genes and the disease type and thus makes diagnosis easier. The feature extraction of such large data is a complex task, and machine learning uses hand-crafted methods for feature extraction. But, by adopting the deep learning technique, it can be possible to infer the important features of genes automatically in microarray data by using various computational tools and methods. Gene expression profiling is very much required to specify the cellular condition concerning critical diseases and medicine development. For a critical disease like cancer, the cancer microarray data have a limited number of samples with large gene expression and features. The expression profile is very large, which makes it a challenging task to classify and predict the feature of high dimension microarray data associated with cancer. Deep learning is the new promising methodology for feature extraction in microarray data and is widely used in the field of genomics, clinical genetics, cancer identification, and diagnosis. In literature, several deep learning methods for averaging, normalized initialization, model selection, and learning algorithms are adopted for prediction of the feature of high dimension microarray data. In this work, the authors have given a detailed description of the gene expression in digital form as well as microarray form and explained the use of it in disease diagnosis. Moreover, a broad literature study has been carried out related to the feature prediction of high dimension microarray cancer data using various deep learning techniques. Last, the work is concluded by giving a critical analysis with several challenges and future research scopes.