Novel Data Mining Methods and Applications in Combinatorial Optimization and Bioinformatics

This text summarizes the identically titled doctoral thesis, which compiles a series of research contributions on the proposal and application of data mining methods in the combinatorial optimization and bioinformatics fields.
It explores traditional methods and, mainly, a novel approach, named MineReduce, for incorporating data mining into metaheuristics, applied to solve various problems. The derived methods achieved relevant results, outperforming state-of-the-art algorithms and finding new best solutions for several benchmark instances of the addressed problems. Additionally, two proposed methods were awarded first place in the MESS 2020+1 metaheuristics competition and second place in the capacitated vehicle routing problem track of the 12th DIMACS Implementation Challenge.
The thesis also reports methodological contributions related to classification and their applications in bioinformatics. It introduces novel approaches to build classifier ensembles for uncertain data, applied for classifying ageing-related genes and predicting drug side effects. The results evidence that the proposed approaches improve the predictive performance of ensembles on uncertain data. Additionally, novel approaches for interpreting Naive Bayes ensembles were introduced and applied to identify relevant features to classify genes, producing consistent results and new insights in this field.