Related Works of Learning with Missing Value
[1] gives comparison of several approaches for solving the missing value problem, including deletion and single imputation techniques. First we can study the characteristic of the missing data. Deletion methods is one of the basic methods to dealing with missing values. In single imputation methods, missing values are filled in and the outcome is a complete data set.
[2] discuss whether missing values should be imputated from the view point of cost-sensitive leaning. The made the claim of “Missing Is Useful” to argue that sometimes it will be better not to imputate missing values.
[3] proposed the method of mixed-attribute imputation to estimate missing values with known values in from different types of attributes.
[4] provides a new angle and solved the MV problem for real big data. They proposed a framework to partition the data and achieve imputation with only a subset of features.
[5] is a paper from Amazon Deutschland. It proposed a method to achieve missing value imputation for non-numerical data.
Methods handling missing values
Deletion Methods
Deleting the whole row or column with missing value, details discussed in [1].
Single imputation methods
Mean imputation
Simply fill with the mean value of the feature.
Regression imputation
Estimate missing feature \(X\) with knowing feature \(Y\). Linear regression or other regression can be used.
Expectation-maximization (EM)
Similar as EM in parameter estimation. Can be combined with regression imputation.
Hot deck methods
Find the nearest records like \(k\)-NN methods.
Ref
[1] Chih-Feng Liu, Thao-Tsen Chen, and Shie-Jue Lee. 2012. A COMPARISON OF APPROACHES FOR DEALING WITH MISSING VALUES. In Proceedings of the 2012 International Conference on Machine Learning and Cybernetics, 15–17.
[2] Shichao; Zhang and Zhengxin Qin. 2005. “Missing Is Useful”: Missing Values in Cost-Sensitive Decision Trees. IEEE TKDE 17, Bd 1 (2005), 80 S.
[3] Xiaofeng Zhu, Shichao Zhang, Zhi Jin, Zili Zhang, and Zhuoming Xu. 2011. Missing value estimation for mixed-attribute data sets. IEEE TKDE 23, 1 (2011), 110–121. DOI:https://doi.org/10.1109/TKDE.2010.99
[4] Pythia Godzilla, Christos Anagnostopoulos, and Peter Triantafillou. 2014. Scaling Out Big Data Missing Value Imputations. In ACM SIGKDD, 651–660.
[5] Felix Bießmann, David Salinas, Sebastian Schelter, Philipp Schmidt, and Dustin Lange. 2018. Deep Learning for Missing-Value Imputation in Tables with Non-Numerical Data. In ACM CIKM, 2017–2025.