Skip to main content

Artificial Intelligence

Artificial Intelligence in project control

The use of Artificial Intelligence methods to improve the forecasting of the final project duration and cost has been investigated in several research studies as described below.

Paper 1. Support Vector Machines for project forecasting

Support vector machine regression has been used widely in the academic literature. In our study, we applied the technique to predict the final time and cost of a project in progress. The paper "Support vector machine regression for project control forecasting" has been published in Automation in Construction.

Paper 2. Nearest neighbour

The nearest neighbour technique is a fairly easy and straightforward technique that can be used to predict the final time and cost of a project. In our paper A nearest neighbour extension to project duration forecasting with artificial intelligence, this technique is tested on a sample of artificial projects.

Paper 2. Comparison of AI methods for project forecasting

The paper "A comparative study of Artificial Intelligence methods for project duration forecasting" is currently under submission in an international journal. In this paper, a number of methods are compared and contrasted by means of a large computational experiment. The methods are implemented in R. In the appendix of the paper, a summary is given on how the training (static phase) and testing (dynamic phase) take place. In order to run the methods on a sample project, the following files need to be downloaded and put into R’s working directory: 

  • OutputProject1.txt.bz2 - this zipped file contains the periodic measurements of the EVM attributes reported in table 1. Each row corresponds with 1 execution, result- ing into 1,000 executions in total. P2 Engine (Vanhoucke (2014)) was employed to generate this file.
  • 5Folds.txt - this file contains the executions that make a division between the training and test set. The 1,000 executions found in OutputProject1.txt.bz2 are partitioned into 800 executions (training set) and 200 executions (test set).
  • 5Folds-Validation.txt - this file contains the executions that make a division between the smaller training set and the validation set. The 800 executions found in the training set are partitioned into 640 executions (training set 2) and 160 executions (validation set).
  • The packages party, gbm and e1071 need to be installed if one wishes to test all AI methods. 
Important note: please ensure that the path of the working directory is changed to the directory in which the required input files reside. The files can be downloaded here.