A Comparison of Machine Learning Methods for Data Imputation
C. Platias, Georgios Petasis
Abstract
Handling missing values in a dataset is a long-standing issue across many disciplines. Missing values can arise from different sources such as mishandling of samples, measurement errors, lack of responses, or deleted values. The main problem emerging from this situation is that many algorithms can’t run with incomplete datasets. Several methods exist for handling missing values, including “SoftImpute”, “k-nearest neighbor”, “mice”, “MatrixFactorization”, and “miss- Forest”. However, performance comparisons for these methods are hard to find, as most research approaches usually face imputation as an intermediate problem of a regression or a classification task, and only focus on this task’s performance. In addition, comparisons with existing scientific work are difficult, due to the lack of evaluations on publicly-available, open-access datasets. In order to overcome the aforementioned obstacles, in this paper we are proposing four new open datasets, representing data from real use cases, collected from publicly-available existing datasets, so as anyone can have access to them and compare their experimental results. Then, we compared the performance of some of the state-of-art approaches and most frequently used methods for missing data imputation. In addition to that, we have proposed and evaluated two new approaches, one based on Denoising Autoencoders and one on bagging. All in all, 17 different methods were tested using four different real world, publicly available datasets.