Landslide Susceptibility Analysis Using Machine Learning Techniques In Penang Island, Malaysia

Gao, Han (2021) Landslide Susceptibility Analysis Using Machine Learning Techniques In Penang Island, Malaysia. PhD thesis, Universiti Sains Malaysia.

Download (728kB) | Preview


Landslides are a natural hazard which cause great losses of lives and properties. Landslide susceptibility analysis (LSA) is of great importance for landslide management and mitigation. This study mainly aims to improve the spatial prediction performance of LSA using machine learning techniques. Since landslide samples account for a small percentage in the raw data, selecting an optimal sample ratio before training machine learning models and increasing the landslide samples in an efficient way are the main research problems. On the one hand, three types of sample ratios are designed to increase the spatial prediction performance through comparative analysis. The equal ratio for datasets is found as the optimal ratio in LSA. Additionally, three oversampling methods, random oversampling technique (ROTE), synthetic minority oversampling technique (SMOTE) and self-creating oversampling technique (SCOTE), are applied to augment the landslide samples. A comparable result is obtained which indicates the efficiency of the augmented landslide samples. Finally, gradient boosting models are developed to integrate with SMOTE and SCOTE in LSA. The area under the curve (AUC) values are considered as the key metric for evaluating the models’ performance. The results show an enhancement in the performance with the highest AUC value of 0.9525. To summarise, the maps produced in this study can provide useful information for the local landslide management and mitigation.

Item Type: Thesis (PhD)
Subjects: Q Science > QA Mathematics > QA1 Mathematics (General)
Divisions: Pusat Pengajian Sains Matematik (School of Mathematical Sciences) > Thesis
Depositing User: Mr Noor Azizan Abu Hashim
Date Deposited: 31 May 2022 17:42
Last Modified: 31 May 2022 17:42

Actions (login required)

View Item View Item