Modelling Landslides Susceptibility by Fuzzy Emerging Patterns

This contribution proposes an approach to model regional landslide susceptibility, based on a supervised learning technique that mines fuzzy emerging patterns on a set of classified data. In our approach the training set contains positive and negative examples of areas, (i.e., slope units), affected or not affected by landslides. The fuzzy emerging patterns characterise the positive and the negative areas exploiting their ability to discriminate between the two classes. The approach consists first, in inducing a set of fuzzy rules, and then in reducing them by retaining those that identify fuzzy emerging patterns for the given training set. The fuzzy rules define the main characteristics of the slope units that are affected or not affected by landslides and are used to classify other slope units in the same region. The classification technique provides an estimate of the hesitation of the decision process, which is a measure of its ability to uniquely associate a slope unit to the susceptible or not susceptible class. In the paper we describe the approach and discuss the preliminary results.