What is CYNERELO?
CYNERELO^{®} is a best-of-breed software solution for time-series forecasting. It accepts qualitative knowledge plus historic data and automatically converts this input into a mathematical forecasting model.
The result does not bother the user with mathematics. However, analysts who are interested in the internal structure can browse the prediction model graphically and even export it for other mathematical toolboxes.
CYNERELO^{®} generates and automatically calibrates a system dynamic prediction model based on a NMPC-graph by means of machine learning. The model is non-probabilistic and a non-trend-extrapolating, but reflects and discloses the causality within the real world scenario.
All input and output data needed by CYNERELO^{®} is ANSI-compliant and can interface easily with office products like MS Excel^{®} as well as with business analytics platforms like those ones from IBM^{®}, Microsoft^{®}, Oracle^{®}, SAP^{®}, and SAS^{®}. CYNERELO^{®} is compatible to modern IT landscapes and prepared to run on client as well as on server side on any Microsoft Windows^{®} platform. CYNERELO^{®} was written in pure ANSI C++ and thus is portable to other operating systems.
Which are the features needed by a high-end forecasting system?
Forecasting is a matter of modelling. If the forecasting model incorporates all essential influences and the model's causal relationships are well reflecting real world, forecasting figures tend to predict reality accurately within a certain range over time.
It is hard to create good forecasting models not only because of lacking knowledge. One important reason for insufficient forecasting models is that features of conventional toolboxes are quite limited in relation to real world’s complexity. When forecasting a time-series, the requirements for model quality are high. Effective forecasting models need to be
multivariate: There is not only one but an arbitrary number of input variables influencing an arbitrary number of output variables.
nonlinear: Many forecasting calculations assume linear causalities because it is much easier to mathematically handle them than nonlinear ones. However, natural systems often are significantly nonlinear.
temporal: The model processes and predicts a time-series. There can be an unknown and varying delay between causes and effects. The model is able to describe growth processes over time.
causal: A web of multiple causalities can be created where causes and effects are distinguishable and may be cascaded and convoluted.
recursive: Effects may influence themselves via causal feedback loops.
intuitive: Analysts should be able to feed the forecasting engine with expert knowledge about the system being forecasted in an easy and human readable way. The forecasting model should be understandable not only by mathematicians and programmers but in fact by everybody whose forecasting subject is of concern.
tolerant: Analysts should not be obliged to specify information almost impossible to know at modelling time, i.e. certain precise mathematical functions and certain mathematical constants like offsets, weights, exponents, and coefficients.
deterministic: The amount and structure of data needed for model calibration should be predictable.
What are CYNERELO’s features in comparison to conventional forecasting methods?