Técnicas de clusterização e comparação de grupos baseados nas características de inovação em empresas de Londrina e região

Abstract

Innovation is often seen as the foundation for success in many areas, such as companies and even countries, but there are difficulties in measuring it. Despite being present in several areas, including in the context of innovation, machine learning (ML) has not yet been used to highlight innovative features in companies. This work aims to employ unsupervised machine learning (ML) techniques for the formation of groups (clusters) of companies in Londrina and region to discuss how the variables related to innovation differ between the formed clusters. For this, a data collection instrument was developed based on CIS 4 (Fourth Community Innovation Survey) created by Eurostat and PINTEC (Industrial Research on Technological Innovation). After creating the collection instrument, a sample of companies in Londrina and region was sent. To extract knowledge from the obtained database, four clustering techniques were used: K-means, K-means with PCA, hierarchical clustering, and hierarchical clustering with PCA. As a result, the companies' distributions in the two clusters created for all algorithms were similar. It was identified that in one of the clusters were allocated companies with a higher degree of innovation, while in the other, companies with a lower degree. Then, analyzes were carried out, with theoretical basis, of three variables in the context of innovation that obtained more differences between clusters formed: regular investment in R&D, partnerships made with different categories to innovate and requests and/or registrations. In the end, it was verified that these variables were considered important for the characterization of the most innovative cluster, however the results indicated that a larger sample should be collected to validate the results obtained initially by this research.