Settlement Prediction Model in Consideration of Static Loading on Soft Clay by Utilising Machine Learning Method

Abstract

The past researchers had deduced that the current condition of marine clay soil in Sabak Bernam, Selangor were highly sensitive soil that is always associated with high settlement and high instability, poor soil properties that are not suitable for engineering requirements. Hence, there is a need to develop an artificial neural network (ANN) model to predict the soil settlement by associating with the current history of soil settlement thus producing a reliable prediction model in the future. The aim of this research is to obtain the best soil settlement  ANN  model which is selected among these four types of soil settlement prediction models (two from deep learning (DL) and two from support vector machine (SVM)) using konstanz information miner (KNIME) machine learning. SVM - dot kernel prediction model exhibited 34% less discrepancy values between measured and predicted Sabak Bernam, Selangor marine clay soil settlement compared to DL - max-out (58%), DL - rectifier (56%) and SVM - neural (59%) and was therefore chosen. Further optimization was made on SVM-dot model in order to reduce the error between measured and predicted value using data splits and performance indexes. At the end of the analysis SVM-dot with root mean square error (RMSE) performance index has achieved further refinement up to 5% between measured and predicted value hence, this model has been chosen and would be suitable in predicting the settlement of problematic soil.