TECHNICAL NOTE Lateral Load Capacity of Piles in Clay Using Genetic Programming and Multivariate Adaptive Regression Spline Pradyut Kumar Muduli • Manas Ranjan Das • Sarat Kumar Das • Swagatika Senapati Received: 9 April 2014 / Accepted: 28 November 2014 Ó Indian Geotechnical Society 2014 Abstract This study presents the development of pre- dictive models of lateral load capacity of pile in clay using artificial intelligence techniques; genetic programming and multivariate adaptive regression spline. The developed models are compared with different empirical models, artificial neural network (ANN) and support vector machine (SVM) models in terms of different statistical criteria. A ranking system is presented to evaluate present models with respect to above models. Model equations are presented and are found to be more compact compared to ANN and SVM models. A sensitivity analysis is made to identify the important inputs contributing to the lateral load capacity of pile. Keywords Lateral loaded pile  Clay  Genetic programming  Statistical method Introduction The design of pile foundation has drawn more attention than other type of foundation structures. The use of axial loaded pile is more frequent and is designed using equa- tions of static equilibrium and other dynamic equations [38]. However, the lateral loaded piles are used in more difficult conditions, particularly in tall and offshore struc- tures. The design of laterally loaded piles is more difficult and requires solution of nonlinear differential equations. The elastic analysis adopting Winkler soil model [38] is not suitable for the nonlinear soil behavior. Matlock and Reese [33] adopted elastic analysis using nonlinear lateral load capacity—deflection (p-y) curves. Portugal and Seco e Pinto [37] used nonlinear p-y curves and finite element method for prediction of the behavior of laterally loaded piles. The above two methods are more accurate and widely used. But, spatial variability of soil is inevitable. Thus, developing a sufficiently accurate site model for FEM analysis requires extensive site characterization effort and desired constitutive modeling of clayey soil is also very difficult, even with considerable laboratory testing. So methods based on field data [8, 27, 34] have become very much popular for the above study, particularly for the preliminary estimate of pile load capacity. These methods are based on pile load test case histories and involve sta- tistically derived empirical equations for estimation of expected lateral load capacity. Artificial intelligence (AI) techniques such as artificial neural networks (ANNs) and support vector machine (SVM) are considered as alternate statistical methods and are found to be more efficient compared to statistical methods [11, 13]. ANN method has been found to be efficient in predicting the pile load capacity in both cohe- sion- less soil and clayey soil compared to traditional P. K. Muduli (&) Department of Civil Engineering, BOSE, Cuttack 753007, Odisha, India e-mail: pradyut.muduli@gmail.com M. R. Das Department of Civil Engineering, ITER, SOA University, Bhubaneswar 751030, Odisha, India e-mail: manasdas.iter@gmail.com S. K. Das Department of Civil Engineering, National Institute of Technology, Rourkela, Rourkela 769008, Odisha, India e-mail: saratdas@rediffmail.com S. Senapati Department of Civil Engineering, Indian Institute of Technology, Madras, Tamilnadu, India e-mail: swagatika.senapati89@gmail.com 123 Indian Geotech J DOI 10.1007/s40098-014-0142-2