Article Preview
TopIntroduction
In recent time, the rising number earthquake has drawn the attention of geotechnical/ earthquake engineers towards the studies of liquefaction susceptibility of soil. Liquefaction causes huge destruction, loss of lives and money due to ground cracking, settlement, tilting, collapsing, overturning of structures such as buildings, dams, etc. It also causes lateral spreading of large areas which result in damage of underground structures such as oil pipelines, optical fibre cables, sewerage systems, port quay walls and bridges. The phenomenon of liquefaction of soil generally occurs when a saturated or partly saturated soil loses almost all of its shear strength and stiffness in reaction to a sudden change in the stress condition such as ground shaking. Different soil behaves differently in response to earthquake. In case of saturated sandy soils, liquefaction occurs instantly. While in the case of clayey soil, particularly that consisting of same size particles i.e. poorly graded and more sensitive soil, may exhibit strain-softening behaviour which is similar to that of liquefied soil, but do not liquefy in the same manner as sandy soils. In liquefaction process the soil spread like a liquid which results in destabilization of the above and surrounding structural features. Till date there is very few method and techniques which can precisely predict the soil liquefaction. So, it is still being a field of great investigation.
There are many research works that had been carried out for liquefaction susceptibility of soils. Seed and Idriss (1967) are among the first to investigate the assessment of liquefaction of soil in which a detail study was accomplished to investigate liquefaction based on shear strength for different criteria. Later, the extension of this investigation was proposed as model by various researchers (Seed and Idriss, 1971, 1982; Seed and Peacock, 1971; Iwasaki, 1978; Iwasaki et al., 1982; Seed et al.,1983; Robertson and Wride, 1998). Further, Artificial neural network (ANN) model was used for the determination of liquefaction susceptibility of soil (Goh, 1994; 1994, 2002; Saka and Ural, 1998; Juang and Chen, 1999; Young-Su and Byung-Tak, 2006). Liquefaction induced lateral displacement was predicted by using Genetic programming (Javadi et al., 2006). Prediction of liquefaction was done by using support vector machine(SVM) model (Samui, 2013). Prediction of liquefaction susceptibility of soil was performed by using extreme learning machine(ELM) (Samui et al., 2016). However, all these models have some limitations. Many problems of geotechnical engineering has been solved by ANN (Yang and Rosenbaum, 2002; Kurup and Griffin, 2006; Yaghmaei-Sabegh and Tsang, 2011). ANN has some limitations such as arriving at local minimum, over fitting problems and slow convergence rate. In case of genetic programming it cannot guarantee optimality and also the solution gets deteriorated as the problem size increases. SVM work very well when we have no idea of the data but the selection of good kernel function is not so easy. In case of ELM, although it is a powerful burden reduction technique, it is found that theoretical performance of ELM depends heavily on the activation function and randomness mechanism (Lin et al., 2015).
In this paper, it is tried to overcome these problems by presenting a comparative analysis between four models namely Fast k-NN model, Naïve Bayes Classifier, Group Method of Data Handling (GMDH) and Decision Forest classification. Further, in fast k-NN, multiple random projection tree (MRPT) model is tested and preferred over the other approximate nearest neighbour search due to high dimensionality of our dataset. This was selected due to its simplicity and easy to implement. All the models were developed in R open source software.