Anzeige
Erschienen in:
2019 | OriginalPaper | Buchkapitel
Feature Based Multivariate Data Imputation
verfasst von : Alessio Petrozziello, Ivan Jordanov
Erschienen in: Machine Learning, Optimization, and Data Science
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
We investigate a new multivariate data imputation approach for dealing with variety of types of missingness. The proposed approach relies on the aggregation of the most suitable methods from a multitude of imputation techniques, adjusted to each feature of the dataset. We report results from comparison with two single imputation techniques (Random Guessing and Median Imputation) and four state-of-the-art multivariate methods (K-Nearest Neighbour Imputation, Bagged Tree Imputation, Missing Imputation Chained Equations, and Bayesian Principal Component Analysis Imputation) on several datasets from the public domain, demonstrating favorable performance for our model. The proposed method, namely Feature Guided Data Imputation is compared with the other tested methods in three different experimental settings: Missing Completely at Random, Missing at Random and Missing Not at Random with 25% missing data in the test set over five-fold cross validation. Furthermore, the proposed model has straightforward implementation and can easily incorporate other imputation techniques.