Evaluation of data mining models for predicting concrete strength

Wong, Chuan Ming (2024) Evaluation of data mining models for predicting concrete strength. Final Year Project, UTAR.

Preview

PDF Download (2221Kb) | Preview

Abstract

Data mining techniques are becoming more popular in recent years due to their abilities to predict any types of data with high accuracy. Conventional techniques for predicting strength of concretes frequently empirical calculations. The efficiency and accuracy of concrete strength prediction can be improved by using data mining techniques. A dataset provided by the Department of Civil Engineering of UTAR that consists of 343 samples of concrete strength along with its mixture proportions is used in this project. The data mining models used in this project are (i) Decision Tree, (ii) AdaBoost, (iii) XGBoost, (iv) Bagging Regressor and (v) Artificial Neural Network. These models are all evaluated with hyperparameter tuning and different feature selection techniques. The feature selection techniques included are (i) Principle Component Analysis, (ii) Boruta and (iii) LASSO. The best performing model is selected and used to generate different sets objective-function that will be selected and used in a Particle Swarm Optimization algorithm to solve a single objective optimization problem that finds the optimal values of each concrete feature to maximize the strength of concrete. The Bagging Regressor model with LASSO is the best performer with a R score of 0.9525. It is selected as the to generate objective�functions for the Particle Swarm Optimization algorithm as it performs performs consistently well with tuned hyperparameters and feature selection. The Particle Swarm Optimization algorithm is able to generate optimal values for the concrete features that maximizes the strength of concrete. The maximum strength that is achievable with the optimal values for each concrete feature found by the Particle Swarm Optimization Algorithm is 27.96.

Actions (login required)