Abstract
<jats:p>This book does not aim to be a comprehensive overview of all possible techniques in the realm of explainability and optimization. It is, conversely, a set of notes oriented towards providing the reader with a compact, yet reasonably flexible set of tools to face a crucial problem: making more effective and better-informed decisions. Together with a subset of standard techniques, these notes also include a few ideas that stemmed from my research in the field. The first section of the book looks at one aspect of this challenge: making the most of the available data, which are already in tabular form. This means being able to treat them and understand them (pre-modeling), derive a model on top of those data understanding its behavior (in-modeling) and ask more specific questions, to a model that may have higher performance but a less interpretable behavior (post-modeling). The second section provides some ideas about how the concepts described in the first section can be adapted and applied even if the input data has a more complex, unstructured form. More specifically, three examples are considered: images, text and graphs. As regards the graphs, i.e., structures with entities (nodes) involved in relations (through connections called edges), they are not only considered as possible input data, but also as one way to represent correlations among input features (by means of a feature graph). The third section is dedicated to a different aspect of the decision-making process: optimization. In fact, there are circumstances in which the aim could be not only pre- dicting and understanding an unknown variable/phenomenon, but making the most out of our decision variables, the variables that we can control. This setting corre- sponds to an optimization problem. We analyze its taxomony, the main algorithms to deal with (different kinds of) optimization problems and a few applications. Among these applications, a wealth management scenario that we use to close the circle: in fact, firstly we optimize the allocation of a given capital, and then we apply one of the techniques described in previous sections (Logic Learning Machine) to make this decision explainable</jats:p>