FaFCNN: A General Disease Classification Framework Based on Feature Fusion Neural Networks

• DOI: 10.48550/arxiv.2307.12518
• Publication Date: 2023-01-01
• Authors: Menglin Kong; Shaojie Zhao; Juan Cheng; Xingquan Li; Ri Su; Muzhou Hou; Cong Cao
• Keywords: Boosting; Robustness; Feature (linguistics); Gradient boosting; Feature Learning

There are two fundamental problems in applying deep learning/machine learning methods to disease classification tasks, one is the insufficient number and poor quality of training samples; another how effectively fuse multiple source features thus train robust models. To address these problems, inspired by process human knowledge, we propose Feature-aware Fusion Correlation Neural Network (FaFCNN), which introduces a feature-aware interaction module feature alignment based on domain adversarial learning. This general framework for classification, FaFCNN improves way existing obtain sample correlation features. The experimental results show that using augmented obtained pre-training gradient boosting decision tree yields more performance gains than random-forest methods. On low-quality dataset with large amount missing data our setup, obtains consistently optimal compared competitive baselines. In addition, extensive experiments demonstrate robustness proposed method effectiveness each component model\footnote{Accepted IEEE SMC2023}.