Angelo Leogrande, Alberto Costantiello, Lucio Laureti, Marco Maria Matarrese
The following article investigates the determinants that lead innovative SMEs to collaborate. Data from 36 European countries is analyzed using Panel Data with Fixed Effects, Panel Data with Random Effects, Pooled OLS, WLS and Dynamic Panel models. The analysis shows that the ability of innovative SMEs to collaborate is positively associated with the following variables: "Linkages", "Share High and Medium high-tech manufacturing", "Finance and Support", "Broadband Penetration", "Non-R&D Innovation Expenditure" and negatively to the following variables: "New Doctorate graduates", "Venture Capital", "Foreign Controlled Enterprises Share of Value Added", "Public-Private Co-Publications", "Population Size", "Private co-funding of Public R&D expenditures". A clustering with kMeans algorithm optimized by the Silhouette coefficient was then performed and four clusters were found. A network analysis was then carried out and the result shows the presence of three composite structures of links between some European countries. Furthermore, a comparison was made between eight different predictive machine learning algorithms and the result shows that the Random Forest Regression algorithm performs better and predicts a reduction in the ability of innovative SMEs to collaborate equal to an average of 4.4%. Later a further comparison is made with augmented data. The results confirm that the best predictive algorithm is Random Forest Regression, the statistical errors of the prediction decrease on average by 73.5%, and the ability of innovative SMEs to collaborate is predicted to growth by 9.2%.