Handling Fluctuating Observability of the Rössler System

Fluctuating observability of the Rössler system may lead to inconsistent prediction accuracy and that may influence causality detection of methods based on predictability improvement. Fluctuating observability is caused by an ill-conditioned numerical problem when drastically changing the magnitude of the observed variable. In this paper, we will show that this problem causes difficulties mainly in cases with a smaller number of data and when using an insufficient model. With enough data and "effort" to find the right model, we can achieve good accuracy of prediction from all variables.