DAEMA: Denoising Autoencoder with Mask Attention

Missing data is a recurrent and challenging problem, especially when using machine learning algorithms for real-world applications. For this reason, missing imputation has become an active research area, in which recent deep approaches have achieved state-of-the-art results. We propose DAEMA (Denoising Autoencoder with Mask Attention), algorithm based on denoising autoencoder architecture attention mechanism. While most use incomplete inputs as they would complete - up to basic preprocessing (e.g. mean imputation) leverages mask-based mechanism focus the observed values of its inputs. evaluate both terms reconstruction capabilities downstream prediction show that it achieves superior performance several publicly available datasets under various missingness settings.