Sinkhole susceptibility mapping: A comparison between Bayes‐based machine learning algorithms

Abstract

Land degradation has been recognized as one of the most adverse environmental impacts during the last century. The occurrence of sinkholes is increasing dramatically in many regions worldwide contributing to land degradation. The rise in the sinkhole frequency is largely due to human‐induced hydrological alterations that favour dissolution and subsidence processes. Mitigating detrimental impacts associated with sinkholes requires understanding different aspects of this phenomenon such as the controlling factors and the spatial distribution patterns. This research illustrates the development and validation of sinkhole susceptibility models in Hamadan Province, Iran, where a large number of sinkholes are occurring under poorly understood circumstances. Several susceptibility models were developed with a training sample of sinkholes, a number of conditioning factors, and four different statistical approaches: naïve Bayes, Bayes net (BN), logistic regression, and Bayesian logistic regression. Ten conditioning factors were initially considered. Factors with negligible contribution to the quality of predictions, according to the information gain ratio technique, were discarded for the development of the final models. The validation of susceptibility models, performed using different statistical indices and receiver operating characteristic curves, revealed that the BN model has the highest prediction capability in the study area. This model provides reliable predictions on the future distribution of sinkholes, which can be used by watershed and land use managers for designing hazard and land‐degradation mitigation plans.

Landslide Susceptibility Assessment by Novel Hybrid Machine Learning Algorithms

“…Machine learning techniques have recently gained good attention among the environmental modeling research community as they are advantageous in efficiently capturing the complex relationship between the environmental predictors and the response, such as flood [33][34][35][36][37][38][39][40][41], wildfire [42], sinkhole [43], drought [44], gully erosion [45,46], groundwater [47][48][49] and land/ground subsidence [27], and landslide in this case [3,13,[50][51][52][53][54][55][56][57]. In due course, researches have also attempted to improve the prediction accuracy and the interpretability of the models through applying various decision-trees machine learning algorithms such as chi-square automatic interaction detector; quick, unbiased and efficient statistical tree [58]; J48 decision trees [59]; ID3 decision trees [60]; random forests [61]; classification and regression trees [62]; alternating decision trees [63]; reduced error pruning trees [3]; naïve Bayes [35,53]; naïve Bayes tree [13,64]; kernel logistic regression [37]; logistic model tree [38,…”

Spatial Prediction of Landslide Susceptibility Using GIS-Based Data Mining Techniques of ANFIS with Whale Optimization Algorithm (WOA) and Grey Wolf Optimizer (GWO)

“…Recent developments of machine learning (ML) have introduced new optimization algorithms, which could be used for optimizing weights for membership function of the neural fuzzy model. Furthermore, ML techniques have recently gained a good attention among environmental modeling research community as they are advantageous in efficiently capturing the complex relationship between the environmental predictors and the response, such as flood [55][56][57][58][59][60][61][62][63], earthquake [64,65], wildfire [66], sinkhole [67], droughtiness [68], gully erosion [69,70], groundwater [71][72][73][74] and land/ground subsidence [75], and landslide in this case [54,59,. Nevertheless, investigation of new optimization algorithms and the neural fuzzy has not been carried out.…”

Application of logistic regression (LR) and frequency ratio (FR) models for landslide susceptibility mapping in Relli Khola river basin of Darjeeling Himalaya, India

“…Many researchers have worked with more than one model and compared to find out which one is most accurate [9]. Recently, machine learning (ML) techniques have become popular in spatial prediction of natural hazards studies such as wildfire [45], sinkhole [91], groundwater and flood [1,12,16,17,40,49,50,59,76,77,82,93], droughtiness [80], gully erosion [7,98], earthquake [4], land/ground subsidence [96] and landslide studies [68,70,79,87,67,97]. ML is a subdivision of artificial intelligence (AI) that uses computer techniques to analyze and forecast information by learning from training data.…”