Selection of most relevant input parameters using WEKA for artificial neural network based concrete compressive strength prediction model

In this paper, a novel approach to predict concrete compressive strength (CCS) at high strength level using artificial neural network (ANN) is proposed. The proposed approach is implemented to train, test and validate using available real 1030 datasets of USI machine learning repository. Data sets utilized to predict CCS included with eight input variables (i.e., blast furnaces slag, cement, fly ash, superplasticizer, water, coarse aggregate, fine aggregate and age) to ANN model which affects the accuracy of CCS prediction. Therefore, the selection of the most relevant input variables to the ANN model is necessary. With this objective, InfoGain Attribute Evaluator with Ranker Search Method using WEKA (a data mining implementation) is applied to find the most relevant input variables. Identified 7 most relevant input variables are used as input to ANN model to predict the CCS. The results obtained validates that the combination of input variables selected through WEKA gives higher prediction accuracy than any other combination of input variables. This method is used to predict the CCS in high strength level.