关键词: 建筑工程, 安全文明施工费, 预测精度, 案例推理技术(CBR), 最大信息系数(MIC), 随机森林(RF)

Abstract:

In order to precisely measure the safety-civilized measure cost of construction projects, the CBR-MIC-RF prediction model method was suggested. 12 influencing parameters of safety and civilization construction cost were chosen as candidate feature variables using the sample data of 61 typical projects collected through a field study. CBR technique was used to retrieve sample similarities in order to build the training sample set for the model. MIC was used to determine the input feature variables of the main model. On this foundation, three Random Forest models (RF, MIC-RF, and CBR-MIC-RF) were merged, and empirical data was used to analyse their prediction precision. The results show that sample similarity search and identification of key feature variables can significantly improve the prediction precision of RF models (mean absolute percentage error(MAPE) is 3.35%). The model prediction precision varies with the similarity thresholds of different levels, and setting the appropriate similarity thresholds is crucial to improving the prediction effectiveness of the models. The CBR-MIC-RF model can achieve a better prediction performance than the support vector machine model.

Key words: construction project, safety-civilized measure cost, prediction precision, case-based reasoning (CBR), maximum information coefficient (MIC), random forest (RF)