•  
  •  
 

Abstract

There are several essential elements in project construction management to be studied appropriately, and priority to these elements, such as cost and duration, is predominantly interesting to be investigated. In this research, the duration of field canal improvement projects (DFCIP) was predicted using two relatively new machine learning (ML) models - the Multivariate Adaptive Regression Spline (MARS) and Extreme Learning Machine (ELM). The targeted DFCIP was calculated using other dependent parameters, such as the length of the pipe, years of construction, the geographical zone of the network, the supplied area with water, and finally the actual cost of the field canal initiative. The development of the model was established using a dataset collected from the open-source literature. The modeling results indicated that MARS over the testing phase attained the least root mean square error (RMSE =6.1609), maximum determination coefficient (R2= 0.7448), mean absolute error (MAE=4.28), a mean absolute percentage error (MAPE=0.05169) with Nash- Sutcliffe Efficiency (Nash= 0.7418) and agreement index (MD= 0.7632). On the other hand, the ELM model achieved RMSE value of 6.869, R2 value of 0.6831, MAE value of 4.7804, MAPE value of 0.05759 with Nash = 0.68108, and MD = 0.73402. These metrics indicate the better performance of MARS over ELM in terms of accuracy and best precision of prediction using all the predictors. The research concluded that the proposed methodology provides reliable technology for duration prediction. In addition, the introduced model can be considered for reliable and robust field canal management.

Download

Included in

Engineering Commons

Share

COinS