Abstarct: Cancer is one of the most common diseases in recent decades has attracted the attention of many researchers in various fields of science. Treatment of this disease using of common therapies is often costly, or fails, or the patient experiences severe side effects. For this reason, the need to develop new therapies is well felt. Genomic variations in DNA cause a variety of cancers in humans. The copy number variations or CNV, as one of a variety of mutations in DNA, has caused various cancers in humans. In order to understand the difference between cancers using of CNVs, in this study, we classify six different classes of cancer in humans using levels of CNV and using of deep learning. In recent years, deep learning has been used to diagnose types of cancer such as lung, skin and breast cancers. Many of these methods have used convolutional neural networks to detect cancer. We use LSTM deep neural network for classification. For this purpose, CNV data from 24174 genes were used as features for building our classifier. Our results of the experiment show an accuracy of 92% in the classification. Then, to analyse the biologically, we identified the genes that had the greatest impact on the development of cancers. For this reason, the maximum dependency and minimum redundancy (mRMR) criterion was used to identify effective genes. Using of mRMR criterion, we identified 200 important genes and then classify these genes with our classifier. The accuracy of more than 81% in the classification indicates the importance of these genes. Finally, we identified 10 important genes that had the most effect on the classification. We examined the genetic function of these genes in order to give statistical results and biological analysis of these genes to the recognition of different types of human cancers and offer suggestions for appropriate treatment for patients.