On Effectiveness of Pre-processing by Clustering in Prediction of C.E. Technological Data with ANNs

Civil Engineering technological data are naturally clustered in a specific way. A black-box model of relation between concrete composition and concrete properties can be constructed using a suitable artificial neural network like Fuzzy ARTMAP that was implemented for the presented experiments. After training the system allows valuable prediction of technological data. It was expected that pretreatment of data by their clustering should enable improved prediction on testing examples. The clustering was realized in two different ways: with k — means algorithm and with GCA approach. The improvement of the precision of predictions was found rather limited, but the final efficiency was better, as more records have been positively recognized. The approach seems to be even more promising in case of data of particular internal structure and application of advanced procedures of clustering.