Prediction of landslide triggers in india using stacked ensemble models

Abstract

Landslides are among the most devastating and recurrent natural disasters, impacting numerous mountainous regions worldwide. India is no exception, with frequent landslides affecting vital economic sectors such as transportation and agriculture, often resulting in significant loss of life. A landslide trigger plays a crucial role in initiating the downward movement of rock, earth or debris on a slope. Triggers include natural events like heavy rainfall, earthquakes and rapid snowmelt, as well as human activities like construction, deforestation and mining, which disrupt slope stability and cause significant damage and loss of life. Accurately identifying the triggers of landslides is essential for reducing risks and mitigating their impact. Many researchers have explored landslide prediction, but only a few have focused on landslide triggers. The main challenge with India’s landslide dataset is its inherent imbalance. To address this issue, the Synthetic Minority Over-sampling Technique (SMOTE) was employed. SMOTE generates synthetic examples for the minority classes, thereby balancing the dataset. This approach improves the model’s ability to correctly identify and predict landslides triggered by less frequent events, leading to more robust and reliable predictions across all classes. To predict landslide triggers, a stacked-based machine learning model has been developed, which outperformed other baseline models with an accuracy of 94.6% on unseen data. This achievement underscores the effectiveness of the stacked ensemble model in capturing the complex relationships between various factors contributing to landslide triggers. The robust performance of the stacked ensemble model highlights its potential as a valuable tool for improving landslide prediction and enhancing public safety.