Abstract
In this work, a novel biclustering-based approach to data imputation is proposed. This approach is based on the Mean Squared Residue metric, used to evaluate the degree of coherence among objects of a dataset, and presents an algebraic development that allows the modeling of the predictor as a quadratic programming problem. The proposed methodology is positioned in the field of missing data, its theoretical aspects are discussed and artificial and real-case scenarios are simulated to evaluate the performance of the technique. Additionally, relevant properties introduced by the biclustering process are also explored in post-imputation analysis, to highlight other advantages of the proposed methodology, more specifically confidence estimation and interpretability of the imputation process.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have