By deconstructing biochemical pathways into intermediary components between genotype and phenotype, gene expression research bridges the gap between DNA information and trait information. These findings provide new paths for uncovering complicated disease genes and biomarkers for illness diagnosis and therapeutic effectiveness and toxicity evaluation. However, the bulk of gene expression data analysis techniques are ineffective for biomarker discovery and illness detection. We proposed BICC framework for cancer classification using transfer learning. The datasets are pre-processed by using decision tree regressor, the features are selected by using Random Forest (RF) with Logistic Regression (LR). The training and validation has done with RESNET50. Finally the classification has done with stacking ensemble classification. The experimental results demonstrate that very high classification accuracy can be attained by hybrid classifier with several biomarkers.