Data mining approach to construction productivity prediction for cutter suction dredgers

Abstract Cutter suction dredgers are important equipment in the dredging engineering. However, dredging construction productivity is affected by many factors such as soil, hydrology, meteorology and underwater sundries, making it difficult to obtain accurate prediction results. This paper presents a new integrated approach for using intelligent data mining algorithms to analyze dredger monitoring data and estimate the effective productivity. Through these combination algorithms, alongside Lasso and Maximal Information Coefficient (MIC), the key features affecting the productivity are filtered out from a 255-dimensional monitoring data set. The continuous mean data cleaning method is proposed according to the characteristics of the filtered data, followed by a clean-up of the feature data to increase smoothness. Four machine learning algorithms, Random Forest, K-Nearest, Naive Bayes and eXtreme Gradient Enhancement (XGBoost), are used to estimate and analyze the productivity of a cutter suction dredger and applied to actual engineering cases. The results show that the prediction accuracy of XGBoost exceeds 90%, which is better than other algorithms used. Finally, the model is compared with the predictive model trained by the characteristics that influence productivity derived from the traditional method. The results still show that the XGBoost model with features obtained from machine learning as input terms has a better prediction effect.