A Machine Learning-Based Approach for Economics-Tailored Applications: The Spanish Case Study

Abstract

The continuous evolution of economy hinders the decision-making process in this field. The former requires sophisticated techniques, and thus, manual and empirical methods are becoming increasingly obsolete. In this paper, we propose a computationally-supported approach that performs an economic profiling of cities based on their economic features and also a prediction of the future evolution of these economical metrics. Our contributions include (I) a data-ingestion module to extract, transform and load data, (II) a profiling module that achieves an unsupervised classification via a new distance-based cellular genetic algorithm and K-means and (III) a prediction unit based on long short-term memory artificial neural networks. Our proposal is tested on Spain, analysing all its 52 cities, where we use 33 types of real-world economic data that have been recorded monthly for fifteen years. All data has been obtained from the Spanish National Institute of Statistics. Our experiments show that the 52 cities could be clustered into only three economic profiles. This decrease in the complexity of entities to be considered allows managers at several levels and countries to take faster and more accurate decisions by dealing with few profiles rather than treating each city apart. Also, we found that each profile contains repetitive similarity patterns that are not only determined by economics but also indirectly ruled by the cities’ geo and demographic situations. Results also showed our prototype’s promising economic predictions.