ETL-CCP: An Effective Ensemble Transfer Learning approach for Cancer Classification and Prediction using Gene Expression data

Abstract

Pan-cancer classification refers to classifying and diagnosing different types of cancer. The deep learning approach enables the detection and diagnosis of cancer across multiple organs and tissue types rather than being limited to a specific type of cancer. In deep learning, transfer learning is a technique whereby a neural network model is first trained on a problem similar to the problem that is being solved. Similarly, Ensemble Learning is a strategy where multiple models are combined to perform a particular task. The ensemble transfer learning method can be an effective alternative for Pan-Cancer classification as it can overcome the limitations of base methods by combining the predictions of multiple models to improve performance. In this work, we have presented a novel ensemble method called ETL-CCP for the classification of 33 cancer types based on gene expression data by combining the predictions of multiple pre-train models of transfer learning to improve performance and robustness and increase interpretability. The method combines the efficiency of DenseNet, ResNet, Inception-V3, and Xception models to capture high-level features from structured input data. In addition to this, a class-weighting mechanism is used to overcome data imbalance issues. The experiments were conducted on a gene expression dataset comprising 10,267 cancer samples from 33 cancer types. Our method achieved a test data accuracy of 96.88%, outperforming current baseline methods. This research demonstrates the potential of ETL-CCP as a powerful tool for cancer detection and highlights the importance of ensemble methods in high-dimensional data analysis.