Weight Predictor Network with Feature Selection for Small Sample Tabular Biomedical Data

Tabular biomedical data is often high-dimensional but with a very small number of samples. Although recent work showed that well-regularised simple neural networks could outperform more sophisticated architectures on tabular data, they are still prone to overfitting tiny datasets many potentially irrelevant features. To combat these issues, we propose Weight Predictor Network Feature Selection (WPFS) for learning from and sample by reducing the learnable parameters simultaneously performing feature selection. In addition classification network, WPFS uses two auxiliary together output weights first layer model. We evaluate nine real-world demonstrate outperforms other standard as well methods typically applied data. Furthermore, investigate proposed selection mechanism show it improves performance while providing useful insights into task.