Landslide Detection with Ensemble-of-Deep Learning Classifiers Trained with Optimal Features

Kumar, A.; Misra, Rajiv; Singh, T. N.; Singh, Vinay Kumar

doi:10.1007/978-3-031-16178-0_21

Cited by 3 publications

(1 citation statement)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Monitoring and measurement of superficial displacement have been used by several researchers in the past and is the easiest way to observe and predict a slope failure activity [22][23][24][25][26][27][28][29][30][31][32][33] Monitoring can also provide useful information to the response of various triggering factors like rainfall, earthquake, freeze & thaw, wind velocity. Although it is a challenging task to fetch physical properties of earth sur-face from real world environment, researchers have proposed various techniques (Table 1) to determine earth surface property (displacement and structural deformation) through continuous monitoring.…”

Section: Literature Reviewmentioning

confidence: 99%

Machine Learning and IoT-Based Approaches to Detect and Predict Rainfall-Triggered Landslides

Kumar,

Singh,

Misra

et al. 2023

RIA

View full text Add to dashboard Cite

Landslides, predominantly triggered by rainfall, pose significant global threats, causing extensive loss of life and severe socio-economic disruptions. Among such calamities, the landslide in Uttarakhand stands as a stark exemplar of the severe repercussions these natural disasters can inflict. This study proposes two sophisticated approaches aimed at the detection and prediction of rainfall-induced landslides. Our initial approach presents a comprehensive analysis of the topographic and hydro-meteorological conditions that catalyzed the catastrophic Kedarnath disaster. This method utilizes an innovative algorithm, validated through machine learning models, in conjunction with an IoT-based application designed to collect critical data necessary for model training and validation. Emphasis is placed on rainfall, identified as a pivotal factor influencing debris flow and lake outbursts during the Kedarnath event. Utilizing the standard deviation of landslide data induced by rainfall from 2013-17, a threshold value was calculated to gauge the severity of such scenarios. The second approach employs a range of machine learning and ensemble learning algorithms to enhance the prediction of rainfall-triggered landslides. These proposed methods were investigated using web-scraped datasets acquired from NASA and IMD portals, with under-sampling and oversampling carried out to mitigate any potential dataset bias. Following extensive exploration and exploitation of diverse learning algorithms, it was inferred that oversampling techniques and the random forest model outperformed alternative models consistently across all performance measures, including Accuracy, Precision, Recall, and F1-Score.

show abstract