Risk Preference Based Support Vector Machine Inference Model for Slope Collapse Prediction

Slope collapse prediction inference errors may be divided into two types, namely 1) predicted collapse followed by actual non-collapse (i.e., α error) and 2) predicted non-collapse followed by actual collapse (i.e., β error). As limited time and information make it difficult to reduce the rate of prediction error, making predictions in a manner that considers decision maker risk preferences in order to consider the preferred α to β error ratio in road slope maintenance strategy formulation represents an important issue. tudy proposes an innovative inference model, the Risk Preference based Support Vector Machine Inference Model (RP-SIM). RP-SIM infers the mapping relationship between input and output variables from historical cases using a Support Vector Machine (SVM), and then uses a fast messy genetic algorithm (fmGA) to conduct an optimal search based on α and β values set in accordance with actual decision maker risk preference.