Predicting Performance Measurement of Residential Buildings Using Machine Intelligence Techniques (MLR, ANN and SVM)

Autor: Salah J. Mohammed, Hesham A. Abdel-khalek, Sherif M. Hafez
Rok vydání: 2021
Předmět:
Zdroj: Iranian Journal of Science and Technology, Transactions of Civil Engineering. 46:3429-3451
ISSN: 2364-1843
2228-6160
DOI: 10.1007/s40996-021-00742-4
Popis: The current study combines three techniques: multi-linear regression (MLR), artificial neural networks (ANN) and support vector machine (SVM) to introduce a novel, alternative approach to using artificial intelligence techniques for earned value management of the construction projects in the Republic of Iraq. The proposed strategies are used to create mathematical models for calculating the Schedule Performance Index (SPI), Cost Performance Index (CPI) and to complete cost performance indicator (TCPI) in residential projects both before and during construction. The MLR technique was used to identify the impact parameters using the statistical package for the social sciences program as well as a web-based software designed to perform the estimating calculations quickly, accurately and without much effort. Besides, the ANN technique was used in the Neuframe software to create new prediction models using the backpropagation algorithm. The SVM is the third technique utilized in the creation of prediction models, whereby the SMOreg algorithm is employed in the WEKA program. In terms of MLR, the values of average accuracy (AA%) were determined to be 95.89%, 96.89% and 95.91% while those of R were 92.911%, 98.916% and 97.837% for each of SPI, CPI and TCPI accordingly. The AA% results of the ANN approach were 83.09%, 90.83% and 82.88%, and the correlation coefficient (R) results were 91.95%, 93.00% and 92.30% for SPI, CPI and TCPI, respectively. Furthermore, SVM findings reveal that the AA% is equal to 94.12%, 71.76% and 84.82% and the correlation coefficient (R) is equal to 99.56%, 91.744% and 99.71% for SPI, CPI and TCPI correspondingly. Finally, the results indicate that the ANN and SVM techniques provide excellent results for estimation when compared to the MLR technique.
Databáze: OpenAIRE
Externí odkaz: