Loess Classification by Region Using Machine Learning Property Values and Reliability Assessment Methods

Abstract

Majority of the loess in Korea is the soil that is formed during rock weathering, and its characteristics vary widely depending on the bedrock type, degree of weathering, and climatic conditions. Consequently, it is necessary to acquire objective data on loess based on its location of production and corresponding properties. Therefore, the purpose of this study is to assess the basic properties of nine representative loess samples. The amount of water required to knead each of these samples was calculated using the standard cement leading method; further, properties including hardness, cohesiveness, springiness, adhesiveness, fracturability, chewi-ness, gumminess and resilience were measured using a texture analyzer. Three machine learning approaches were adopted to classify the samples based on the measured features, namely: support vector machine (SVM), k-nearest neighbors (KNN), and deep neural network (DNN). The results showed that SVM produced the highest training and test classification accuracy; meanwhile, DNN presented the best generalization ability.