Extending a fuzzy inductive reasoner with classification procedures
Danny Van Welden, Ejh Kerckhoffs, Gc Vansteenkiste · Ghent University Academic Bibliography (Ghent University) · 1998
SAPS (System Approach Problem Solver) is a fuzzy inductive reasoner that relies on a pattern recognition approach for qualitative system identification. It was initially implemented by H. Uyttenhove in APL, (Uyttenhove 1978), and then reimplemented as a toolbox within the framework of the CTRL-C environment (and later MATLAB) by F. Cellier, (Cellier 1987). SAPS searches a kind of "mental" model for dynamic directed black-box systems with an underlying mainly deterministic relationship between outputs and inputs. A candidate time-invariant pattern "flattens" the input-output data into the state-observation space, such that the state-observation records are now static. Taking furthermore, the idea of starting from the most complex pattern (later to be defined) one obtains a stream of data records that leans itself perfectly for classification. Hence, the pattern recognition approach in SAPS can be situated in the supervised learning paradigm: outputs of a system under investigation form the responses known at each time instance. As a consequence, different data-mining methods can now be applied to find patterns in the state-observation matrix. This may constitute a new approach to finding patterns in temporal databases.