Using heavy-tailed distributions to stress-test kernel methods for segregating the firms that are likely to survive

While kernel-based learning methods have emerged during the last two decades as major tools to effectively manage uncertainty, heavy-tailed distributions remain a major challenge for modelers who aim to predict the future behavior of complex systems. In this article, Weibull distribution has been used to stress-test kernel-based methods and study more specifically the impact of heavy-tailed distributions on the performance of Fisher kernels in identifying the potential for collapse of an enterprise based on its stock price.