PREDICTION OF SLOPES SAFETY FACTOR USING ANN AND TREE-BASED MODELS

Abstract

Slope stability assessment is a non-linear engineering problem. The complex relationship between factors that affect slope instability is challenging to investigate theoretically and mathematically. This paper proposes the prediction of slopes safety factor, subjected to circular failure mode using artificial neural network (ANN) and two tree-based models using M5P and random forest (RF) algorithm. A dataset including 46 slope cases was divided into a 70/30 ratio to train and validate the model. The input parameters for slope stability evaluation include slope angle, slope height, cohesion, internal friction angle, unit weight and pore pressure ratio. The corresponding output parameter is a factor of safety (FS). The finest ANN structure was obtained as 6-7-1 using the trial-and-error method. The proposed model chooses elliot as the best activation function at a learning rate and momentum of 0.7 and 0.5 for the accurate estimation of FS. In validation phase of ANN, M5P and RF models, the coefficient of determination (R2) results as 0.94, 0.86 and 0.81, respectively. The result suggests that the developed ANN model provides higher accuracy than tree-based models which can be an effective tool for slope stability assessment. Furthermore, the significance of input variables for FS prediction has been investigated through sensitivity analysis, which identifies the complex relationship between the input and output variables. The accuracy achieved from the prediction model is crucial for preventing the risk of slope failure and increasing the slope safety during preliminary design stage.