TRAINING KSIM MODELS FROM TIME-SERIES DATA

The suitability of KSIM models derived from group participation strate gies is critically evaluated in a comparison with models generated by a gradient descent learning algorithm. Two learning algorithms are des cribed to train KSIM socioeconomic models. The algorithms are used to train KSIM cross-impact matrices from initially random weights to fina l values producing a model that will closely fit a given time series. The time series can be obtained by integrating a KSIM model or by usin g raw data from other sources. KSIM modeling previously relied on insi ght, intuition, or knowledge of KSIM modeling to find suitable paramet ers. The training algorithms provide an organized approach to the mini mization of a suitable cost function. At the same time, any system kno wledge can be incorporated into initial conditions with learning perfo rmed around solid physical foundations. Some limits of the dynamic per formance of the KSIM model are noted, further establishing the unsuita bility of the KSIM model for many real systems.