Improving algorithmic optimisation method by spectral clustering

In this paper, a spectral algorithm for effort estimation is evaluated. As effort prediction method the Algorithmic Optimisation Method is employed. Spectral clustering is used in version of normalized Laplacian matrix and k-means algorithm is used for clustering eigenvectors. Results shows that clustering lowers a Mean Absolute Percentage Error by 6% and Sum of Squared Errors/Residuals is decreased by 43,5%. Difference in mean value of residuals is statically significant (p = 0.0041, at 0.05 level). © Springer International Publishing AG 2017