Optimizing of predictive performance for construction projects utilizing support vector machine technique
Construction projects still face the old–new problem of delivering the projects within the predefined time and cost. This problem becomes more complicated with when addendums and variations are considered during the projects. This study aimed at developing an artificial intelligent model using suppo...
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
Taylor & Francis Group
2019
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/1dac06763cc848f4aebca1fb00fc77f8 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Construction projects still face the old–new problem of delivering the projects within the predefined time and cost. This problem becomes more complicated with when addendums and variations are considered during the projects. This study aimed at developing an artificial intelligent model using support vector machine (SVM) technique to predict the time and cost indices of projects. Data from 21 tunnel projects implemented in Kurdistan, Iraq were collected and used in this study. The input data include five variables, namely contract value, contract duration, number of change orders, number of conflicts, and classification of company. WEKA––a set of software for machine learning and data mining––developed at the University of Waikato in New Zealand was used to build SVM model to predict the time and cost indices. The collected data were split by default into a training set of 65%, a testing set of 10% and a validation set of 25%. The results show that SVM model I successfully predicted the cost index not only for the trained data, but also for projects with input parameters out of the range of the training inputs. Mean Absolute Percentage Error (MAPE) and Average Accuracy (AA) for SVM prediction of cost index were found to be 13.9% and 86.1%, respectively. The SVM model II accurately predicted the time index with MAPE and AA of 3.4% and 96.6%, respectively. |
|---|