Data-driven models for uniaxial compressive strength prediction applied to unseen data

Data Mining (DM) techniques have been successfully applied to solve a wide range of real-world problems in different real-world domains, particularly in the field of geotechnical civil engineering. A remarkable example is their use in Jet Grouting (JG) technology. Due to the high number of parameters involved and to the heterogeneity of the soil, JG mechanical properties prediction, as well as columns diameter, are complex tasks. Accordingly, the high learning capabilities of DM, namely of the Support Vector Machine (SVM), were applied in the development of new approaches to accurately perform such tasks. This paper aims to assess the SVM model performance trained to predict Uniaxial Compressive Strength (UCS) of JG samples extracted directly from JG columns, when applied to a new set of records collected from a new JG work not contemplated in the database used during the model learning phase. The achieved results highlight the importance of the model domain applicability, as well as the restrictions and recommendations for its generalization when applied to new JG work data not contemplated in the training dataset.The authors wish to thank to Fundacao para a Cienciae a Tecnologia (FCT) for the financial support under the Pos-Doc grant of strategic project PEstOE/ECI/UI4047/2011. Also, the authors would like to thank the interest Tecnasol-FGE company for providing all data needed