Cancer is currently one of the most threatening diseases. The analysis of high dimensional microarray data is a method that is frequently used for cancer diagnosis. Multi category imbalance data analysis in the medical sciences plays crucial role. The growing number of cancer patients has made the necessity for cancer classification absolutely necessary. In this study, PE-SU-R-CNN and PE-ChS-R-CNN are employed as parallel ensemble feature selection based deep learning classifiers to address issues for multi-class cancer classification. The deep learning convolutional neural network with soft-max activation function at decision layer is used. The most well known nine cancer microarray gene expression datasets are used in the experiments. The proposed system's empirical findings are compared with deep learning techniques based on bio-inspired feature selection firefly and elephant searches. Finally, a one-way ANOVA statistical significance test with a post hoc Tukey's test is used to draw several inferences about the optimal classifier model to use. The proposed parallel model shows significant improvement over existing methods.