Missing Value Imputation With Unsupervised Backpropagation

Unfortunately, real-world datasets often include only samples of observed values mixed with many missing or unknown elements. Missing values may occur due to human impatience, human error during data entry, data loss, faulty sensory equipment, changes in data collection methods, inability to decipher handwriting, privacy issues, legal requirements, and a variety of other practical factors. Thus, improvements to methods for imputing missing values can have far-reaching impact on improving the effectiveness of existing learning algorithms for operating on real-world data. We present a method for imputation called Unsupervised Backpropagation (UBP), which trains a multilayer perceptron (MLP) to fit to the manifold represented by the known features in a dataset. We demonstrate this algorithm with the task of imputing missing values, and we show that it is significantly more effective than other methods for imputation. Backpropagation has long been a popular method for training neural networks (Rumelhart et al., 1986; Werbos, 1990).