A Revealed Imperfection in Concept Drift Correction in Metabolomics Modeling

Prediction models that rely on time series data are often affected by diminished predictive accuracy. This occurs from the causal relationships of the data that shift over time. Thus, the changing weights that are used to create prediction models lose their informational value. One way to correct this change is by using concept drift information. That is exactly what prediction models in biomedical applications need. Currently, metabolomics is at the forefront in modeling analysis for phenotype prediction, making it one of the most interesting candidates for biomedical prediction diagnosis. However, metabolomics datasets include dynamic information that can harm prediction modeling. The study presents concept drift correction methods to account for dynamic changes that occur in metabolomics data for better prediction outcomes of phenotypes in a biomedical setting.