on the hydro-chemical properties alone of collected water samples. Considering the
numerical predictors, it is also evident that various linear regression machine learning techniques were applied. However, when it comes to a multidimensional case
(Podgorski et al. 2018), the preferred method is logistic regression, a binary classification technique.
4.6 Application of AI in Predicting Nitrate Contamination
of Groundwater
Only a few studies on prediction of nitrate contamination of groundwater use AI
techniques (Table 4.6). In one such study, Sajedi-Hosseini et al. (2018) developed
boosted regression tree, multivariate discriminant analysis, and support vector
machine models based on the hydrologic and topographic parameters of 102 water
samples collected from Iran.
In a recent study, the authors developed random forest, classification and regression trees, and support vector machine models using hydrogeological, hydrological,
census, and Normalized Difference Vegetation Index (NDVI) data extracted from
remote sensing images for 110 water samples collected in Spain (Rodriguez-Galiano
et al. 2018). In another, the authors developed a random forest regression model
examining the land use, soil type, hydrogeology, topography, climatology, type of
region, and nitrogen fertilizer application factors of 250 groundwater samples
collected in Africa (Ouedraogo et al. 2019). Comparatively, more comprehensive
fluoride prediction models have been developed by Messier et al. (2019); these apply
random forest, gradient boosted machine, support vector machine, neural network,
and kriging techniques to a larger dataset of 22,000 samples collected from private
wells in the U.S. state of North Carolina, and they consider nearly 120 variables
(Table 4.5). Additionally, Knoll et al. (2019) have developed multiple linear regression, classification and regression trees, random forest, and boosted regression tree
models on 1890 groundwater samples collected from the German state of Hesse. The
authors employed the hydrogeological, hydrologic, land use, and soil properties of
the sampled sites (Knoll et al. 2019).
It is evident that nitrate prediction models have been developed using various
hydrogeologic, land use, and topographic properties, and that the random forest
method has been the most preferable option.
4.7 Discussion and Conclusions
According to the best of the knowledge gained from the literature review, certain
artificial intelligence algorithms have been suggested and developed for classification tasks related to the mapping of environmental problems. According to their
98
S. K. Singh et al.
numerical predictors, it is also evident that various linear regression machine learning techniques were applied. However, when it comes to a multidimensional case
(Podgorski et al. 2018), the preferred method is logistic regression, a binary classification technique.
4.6 Application of AI in Predicting Nitrate Contamination
of Groundwater
Only a few studies on prediction of nitrate contamination of groundwater use AI
techniques (Table 4.6). In one such study, Sajedi-Hosseini et al. (2018) developed
boosted regression tree, multivariate discriminant analysis, and support vector
machine models based on the hydrologic and topographic parameters of 102 water
samples collected from Iran.
In a recent study, the authors developed random forest, classification and regression trees, and support vector machine models using hydrogeological, hydrological,
census, and Normalized Difference Vegetation Index (NDVI) data extracted from
remote sensing images for 110 water samples collected in Spain (Rodriguez-Galiano
et al. 2018). In another, the authors developed a random forest regression model
examining the land use, soil type, hydrogeology, topography, climatology, type of
region, and nitrogen fertilizer application factors of 250 groundwater samples
collected in Africa (Ouedraogo et al. 2019). Comparatively, more comprehensive
fluoride prediction models have been developed by Messier et al. (2019); these apply
random forest, gradient boosted machine, support vector machine, neural network,
and kriging techniques to a larger dataset of 22,000 samples collected from private
wells in the U.S. state of North Carolina, and they consider nearly 120 variables
(Table 4.5). Additionally, Knoll et al. (2019) have developed multiple linear regression, classification and regression trees, random forest, and boosted regression tree
models on 1890 groundwater samples collected from the German state of Hesse. The
authors employed the hydrogeological, hydrologic, land use, and soil properties of
the sampled sites (Knoll et al. 2019).
It is evident that nitrate prediction models have been developed using various
hydrogeologic, land use, and topographic properties, and that the random forest
method has been the most preferable option.
4.7 Discussion and Conclusions
According to the best of the knowledge gained from the literature review, certain
artificial intelligence algorithms have been suggested and developed for classification tasks related to the mapping of environmental problems. According to their
98
S. K. Singh et al.