Landslide Susceptibility Assessment Using an AutoML Framework.

Area of Science:

• Geosciences
• Environmental Science
• Data Science

Background:

• Landslide risks are escalating globally, causing significant personal and material losses.
• Effective landslide risk mitigation requires enhanced hazard identification and understanding through susceptibility assessment.
• Machine learning (ML) is increasingly used for landslide susceptibility, leveraging diverse data sources.

Purpose of the Study:

• To develop and validate a novel landslide susceptibility assessment methodology using an automated machine learning (AutoML) framework.
• To compare the performance of 16 ML algorithms within the AutoML framework for landslide prediction.
• To demonstrate the advantages of AutoML in facilitating ML model development for landslide hazard analysis.

Main Methods:

• Development of a landslide susceptibility assessment methodology based on an AutoML framework.
• Application of the AutoML framework to the center and southern region of Guerrero, Mexico.
• Comparative performance evaluation of 16 distinct machine learning algorithms, including Extra Trees.

Main Results:

• The Extra Trees algorithm achieved the highest performance with an Area Under the Curve (AUC) of 0.983.
• The AutoML framework demonstrated superior results compared to other similar landslide susceptibility methods.
• The methodology facilitated deeper data analysis and understanding of landslide-influencing variables.

Conclusions:

• The proposed AutoML framework offers a robust and efficient approach to landslide susceptibility assessment.
• This methodology enhances landslide prediction accuracy, crucial for effective disaster risk management.
• AutoML frameworks enable researchers to focus on data interpretation and gain insights into landslide processes.