Classification algorithms in the material science and engineering data mining techniques

Abstract

Data mining is an analytical process, which deals with the study of large data sets in search of patterns, correlations between data, and later their evaluation. The goal of data mining is usually prediction, among others sales volume, customer activities, extension ratios or the scale of customer loss. Data mining techniques allow finding previously unknown dependencies and schemas that can be used to support decision making or database description. Data mining techniques are developing very quickly and are more and more often used not only in typical fields such as customer relationship or management, but also in medicine, biomechanics, industry, materials sciences or mechanical engineering. The aim of this work is to evaluate the effectiveness of selected data mining techniques for predicting the concrete compressive strength, and to identify the features having the greatest impact on its compressive strength. The study analyzed the data of 1030 concrete samples using five known classification algorithms (C4.5, Random Forest, Naive Bayes Classifier, Supporting Vector Machine SVM) and neural networks (Multilayer Percepton), which allowed to build an exploration model given with an accuracy of over 99%. Potential features of concrete that may affect its compressive strength are also pointed out.