70 Assessment of Prediction of Weathering Degrees for Drilling Core According to Samples Color

In this paper, we study a real multi-classification problem. The goal is to predict the weathering degrees of drilling core samples in situ, on the basis of their color. These weathering degrees (6 classes) are provided by an expert (geologist) and validated by chemical analysis. Our objective is to build a classifier making possible the prediction of the weathering degrees according to the color parameters measured with a spectrophotometer. After choosing the colorimetric system, we propose a features selection method. Then, we build two classifiers: one based on a perceptron multi-layer neural network (NN) and the other based on the fuzzy k-nearest-neighbor (knn) methods. Comparison of results between the methods used and other classical ones are given.