Application of Machine Learning Algorithms to Seismic Energy Dissipation of Rocking Foundations during Earthquake Loading

The objective of this study is to develop preliminary predictive models for seismic energy dissipation of rocking foundations using machine learning algorithms and supervised learning technique. Data from a rocking foundation database consisting of dynamic base shaking experiments conducted on centrifuges and shaking tables have been used for the development of multivariate linear regression (MLR) model using stochastic gradient descent optimization and weighted k-nearest neighbors (k-NN) model. Normalized seismic energy dissipation in soil during rocking is considered as the performance parameter and the input features include critical contact area ratio and rocking coefficient of soil-foundation system, and peak ground acceleration and Arias intensity of earthquake. It is found that though both MLR and weighted k-NN models perform satisfactorily in capturing the complex relationships between seismic energy dissipation and input features of rocking systems, weighted k-NN model consistently outperforms MLR model for the problem considered.