Predicting ENR's Construction Cost Index Using the Modified K Nearest Neighbors (KNN) Algorithm

摘要

Construction cost index (CCI) is calculated monthly and published by Engineering News Record (ENR). It is the weighted average price of four major construction components from twenty cities in the U.S., and it generally represents the level of cost for the construction industry. It has, therefore, been widely used for cost estimation, bid preparation, and investment planning. Cost estimators and investment planners, however, are not only interested in the CCI at present, but are also interested in forecasting and predicting the value of CCI in near future. Although CCI increases over the long term, it is subject to considerable short-term volatilities, which make it problematic for cost estimators to accurately bid for construction projects. The accurate prediction of the construction cost can result in more accurate bids and avoid underestimation or overestimation. This paper presents a method that utilizes economic variables to predict CCI by applying the modified K nearest neighbors (KNN) algorithm. It first identifies several economic variables that are highly correlated to CCI and collects the data for the period from 1994 to 2013. Then it splits the data into two parts: the first 95% of the data as the training set and the last 5% of the data as the test set. Finally, prediction models are developed by the training sets and tested on the test set to measure the prediction performance. Experimental results show that the modified KNN algorithm yields very small prediction error under the given dataset. It is anticipated that the proposed method will be useful for predicting CCI and several other construction time series indices in the short term.