Active-Learning Approaches for Landslide Mapping Using Support Vector Machines

Ex post landslide mapping for emergency response and ex ante susceptibility modelling hazard mitigation are two important application scenarios that require the development of accurate, yet cost-effective spatial models. However, manual labelling instances training machine learning models is time-consuming given data requirements flexible data-driven algorithms small percentage area covered by landslides. Active aims to reduce costs selecting more informative instances. In this study, common active-learning strategies, uncertainty sampling query committee, combined with support vector (SVM), a state-of-the-art machine-learning technique, in case study order assess their possible benefits compared simple random locations. By “informative” instances, SVMs active based on outperformed both query-by-committee strategies when considering mean AUROC (area under receiver operating characteristic curve) as performance measure. Uncertainty also produced stable performances smaller standard deviation across repetitions. conclusion, limited conditions, reduces amount expert time needed SVM training. We therefore recommend incorporating into interactive workflows, especially settings, but modelling.