Sandra Benítez, Universidad de Sevilla
New models for enhancing data knowledge. A Mathematical Optimization outlook.
We are living in the era of Data Science and Big Data. It is changing the way doctors treat or diagnose diseases,how businesses anticipate what their customers want, or even how governs can predict and thereby ease the effects of natural disasters. Nevertheless, in order to do this in the most effective way, collection and the posterior understanding of data is necessary.
In Data Science, the main purpose is to extract knowledge from datasets. In this framework, our contribution is differenciated in two main parts, and both of them will be considered during the talk. First, from a general point of view of classification, we make use of Support Vector Machines. Here, I will explain a succession of works in which we make an intensive use of such supervised learning algorithm, which is in fact one of the most popular and powerful classification and regression methods in the literature. In those works, we take advantage of the flexibility that Mathematical Optimization provides. Such a flexibility allows us to modify and therefore improve the performance of the algorithm. This is done by simple modifications of the formulation, such as the incorporation of new variables or constraints. Secondly, now from a Business perspective, we have modified the model of Data Envelopment Analysis, in order to perform simultaneously Benchmarking and Feature Selection in both the inputs and the outputs, also under the possible consideration of a DMU clustering, adding a game theory insight to the results we obtain. In all these works we handle both the modelling of Mathematical Optimization formulations as well as its implementation in software and its solution using efficient avant-garde solvers.
All the different methods explained during the talk will be illustrated by a number of applications in real-world data.