A data-driven missing value imputation approach for longitudinal datasets

Abstract Longitudinal datasets of human ageing studies usually have a high volume missing data, and one way to handle values in dataset is replace them with estimations. However, there are many methods estimate values, no single method the best for all datasets. In this article, we propose data-driven value imputation approach that performs feature-wise selection method, using known information rank five selected, based on their estimation error rates. We evaluated proposed two sets experiments: classifier-independent scenario, where compared applicabilities rates each method; classifier-dependent predictive accuracy Random Forest classifiers generated prepared baseline doing (letting classification algorithm internally). Based our results from both experiments, concluded generally resulted models more accurate estimations data better performing classifiers, longitudinal ageing. also observed devised specifically had very This reinforces idea temporal intrinsic worthwhile endeavour machine learning applications, can be achieved through approach.