Landslide assessment of the Strača basin (Croatia) using machine learning algorithms

Miloš Marjanović (Author), Miloš Kovačević (Author), Branislav Bajat (Author), Snježana Mihalić Arbanas (Author), Biljana Abolmasov (Author)

Abstract

In this research, machine learning algorithms were compared in a landslide-susceptibility assessment. Given the input set of GIS layers for the Starča Basin, which included geological, hydrogeological, morphometric, and environmental data, a classification task was performed to classify the grid cells to: (i) landslide and non-landslide cases, (ii) different landslide types (dormant and abandoned, stabilized and suspended, reactivated). After finding the optimal parameters, C4.5 decision trees and Support Vector Machines were compared using kappa statistics. The obtained results showed that classifiers were able to distinguish between the different landslide types better than between the landslide and non-landslide instances. In addition, the Support Vector Machines classifier performed slightly better than the C4.5 in all the experiments. Promising results were achieved when classifying the grid cells into different landslide types using 20% of all the available landslide data for the model creation, reaching kappa values of about 0.65 for both algorithms.

Keywords

landslides;support vector machines;decision trees classifier;Starča Basin;

Data

Language:	English
Year of publishing:	2011
Typology:	1.01 - Original Scientific Article
Organization:	UM FGPA - Faculty of Civil Engineering, Transportation Engineering and Architecture
UDC:	550.348.435(497.5)
COBISS:	262545920
ISSN:	1854-0171
Parent publication:	Acta geotechnica Slovenica
Views:	647
Downloads:	50
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	Slovenian
Secondary title:	Bočna nosilnost kratkih togih pilotov v dvoslojnih nevezljivih tleh
Secondary abstract:	V tej raziskavi so avtorji primerjali algoritme strojnega učenja v okviru prognoze drsenja terena. Na osnovi GIS slojev področja kotline Starča, ki so vključevali geološke, hidrogeološke, morfometrijske in druge prostorske podatke, je napravljena klasifikacija mrežnih celic na (i) primerih »drsečega« in »stabilnega terena«, (ii) različnih tipih drsečega terena (»potencialen-neaktiven«, »stabiliziran-saniran« in »reaktiviran«). Po optimizaciji parametrov modela za C4.5 decision trees in Support Vector Machines so primerjali dobljene rezultate klasifikacije s pomočjo kappa statistike. Rezultati kažejo, da sta omenjena modela bolje razlikovala med različnimi tipi drsečega terena kot med drsečim in stabilnim terenom. Prav tako je bil klasifikator Support Vector Machines v vseh preizkusih nekoliko uspešnejši od C4.5. Spodbudne rezultate so dobili v eksperimentu, kjer so klasificirali različne tipe drsečega terena, uporabili pa so samo 20% od skupnega števila podatkov o drsečem terenu. V tem primeru so za oba klasifikatorja dobili vrednost kappa okoli 0.65.
Secondary keywords:	geofizika;plazovi;modeli;algoritmi;kotlina Strača;Hrvaška;
URN:	URN:SI:UM:
Type (COBISS):	Scientific work
Pages:	str. 44-55
Volume:	ǂVol. ǂ8
Issue:	ǂ[no.] ǂ2
Chronology:	2011
ID:	10941017