An Innovative Perspective on Metabolomics Data Analysis in Biomedical Research Using Concept Drift Detection

One of the most challenging scenarios of data analysis is prediction using time series data. As the underlying causal relationships of the data shift over time, a classification model trained on data at earlier points within the course starts to yield incorrect predictions on the current data. This phenomenon in machine learning is called concept drift. Within biomedical data, one of the molecular networks that changes significantly over a time is the metabolome. Using metabolomics analysis in biomedical applications produces an ideal tool in preventive healthcare, the pharmaceutical industry, and even ecology engineering. This study provides an innovative perspective on the analysis of metabolomics datasets using the concept of drift detection. The evaluation is based on two main objectives. The first objective is connected to the concept drift detection in available metabolomics datasets, and the second objective is to provide the assessment of commonly used machine learning tools for the best general detection approach in metabolomics datasets. The application of concept drift to metabolomics data has never been carried out before and is an original take on the analysis of highly dynamic molecular networks.