Keywords: data mining, k nearest neighbour, principle component analysis, concrete, compressive strength, prediction accuracy.

Concrete is one of the most widely used structural materials of construction industry. Traditionally, concrete has been fabricated from a few well-defined components: cement, water, fine aggregate, coarse aggregate, etc. The compressive strength of concrete is generally regarded as its most important property, and many other physical properties of concrete, such as elastic modulus, water tightness or impermeability are derived from concrete strength, as generally there exists a direct relationship between concrete strength and these parameters.

In this paper an instance based learning and a multivariate data reduction method are combined to provide a model for estimation of the 28-day compressive strength of concrete using a laboratory tested concrete mix data set. Concrete mix proportioning parameters such as water cement ratio, fine aggregate percentage, coarse aggregate percentage, water content, cement and admixtures content are used as input of the multivariate model.

This paper demonstrated that it is feasible to use k-nearest neighbour (KNN) instance-based learning in the estimation of 28-day strength of concrete. The fundamental idea of the KNN algorithm is to search for analogs of a feature vector (vector of variables for which analogs are sought) based on similarity criteria in the observed data [1]. The efficiency of the proposed method is verified using an actual laboratory tested concrete mix proportioning dataset. The input data consisted of eight concrete mix ingredient data. Since the ingredient data is highly collinear, a principal component analysis, a multivariate data compression method, was also applied to input data to obtain a new uncorrelated orthogonal input data [2].

KNN experiments were performed on the original data and principal components for different number of neighbours and results indicated that for both cases using four neighbours yields the best results. The RMSE obtained by combining the KNN and the PCA were lower than when the original data was used. This is important as this is achieved by using fewer variables in the KNN which could lead to saving valuable computational time when we are dealing with larger data sets.

Generally, the results indicated that the proposed methods may be used as very efficient tools in the prediction of concrete strength. The estimation problem considered in this article is typical of many scientific and engineering data sets, and the proposed method could be useful in many other applications.

1

D. Hand, H. Mannila, P. Smyth, "Principles of Data Mining", The MIT Press, 2001.

2

A.A. Afifi, V. Clark, "Computer-aided multivariate analysis", Chapman & Hall, New York, 1996.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)