Groundwater Flow Rate Prediction From Geo‐Electrical Features Using Support Vector Machines

Abstract

Water is a finite resource, unequally distributed over time and space and the use of surface water as an auxiliary source of water supply requires many treatments at exorbitant costs. In developing countries, groundwater is the best source of drinking water outside of anthropic pollution (

Ensemble learning paradigms for flow rate prediction boosting

“…However, the prediction of FRs using geo-electrical features seems novel in the literature. The first work was published using the SVMs as predictors with a global score on validation equal to 70% of correct prediction (Kouadio et al 2022). At a glance, this rate is satisfactory in terms of minimizing unsuccessful drillings.…”

“…Secondly, the GRAN itself is known as an unproductive structure especially the wide fracture found on that structure (Lasm 2000). Based on the Baseyian computation of geoelectrical features in Kouadio et al (2022), the authors demonstrate that wide-fractures in that area have a 6.07 % chance to obtain groundwater contrary to the narrow-fracture (82.35%). This confirms the geological complexity of the survey area (Gnamba et al 2014).…”

“…Whilst, the ordinary ML approaches learn one hypothesis from the training data (Zhou 2007). The approach introduced in this workflow first entails using electrical features defined from the ERP and the VES by Kouadio et al, (2022) such as type, shape, magnitude, power, pseudo-fracturing index, and ohmic-area additional to the geology of the area to compose the geo-electrical features. Secondly, the observed FRs such as the unproductive boreholes (FR0) composed of the dry boreholes (𝐹𝑅 = 0 𝑚 3 ℎ𝑟 ⁄ ), the unsustainable boreholes ( 0 < 𝐹𝑅 ≤ 1 𝑚 3 ℎ𝑟 ⁄ ) and the productive boreholes FR1 (𝐹𝑅 > 1 𝑚 3 ℎ ⁄ 𝑟) are categorized into two classes.…”

“…As a consequence, a lot of investments from the governments and funders as well as the profits for geophysical and drilling companies have been lost (Bjornlund 2004;Xepapadeas and Koundouri 2004). To address this problem while using ERP and VES, with a reasonable amount of data, we proposed an ensemble machine learning (EML) method composed of four base learners (BLs) algorithms: (i) the Logistic Regression (LR) (Cox, 1958, Geron, 2019Raschka and Mirjalili, 2019), (iii) K-nearest neighbors (KNN) (Altman, 1992;Grus, 2015), (ii) Decision Tree (DT) (Ho, 1995;Wu et al, 2008;Nath & Levinson 2014, Breiman 2019, and Yariyan et al, 2020 and (iv) the pretrained Support Vector Machines (pSVM) of Kouadio et al (2022). The latter corresponds to the Gaussian radial basis function (RBF) kernel of SVMs (Vapnik 1998;Chen et al 2006;List and Simon 2009;Chang and Lin 2011;Murphy 2012;Xu et al 2014).…”

“…𝑚) is the training vectors, and m ∈ ℕ is the set of examples. The dot product 𝜙 𝑙 (𝒙 𝒊 ) 𝑇 𝜙 𝑙 (𝒙 𝒋 ) is called the kernel trick and is based on the original vectors 𝒙 𝒊 and 𝒙 𝒋 (Cristianini and Shawe-Taylor 2000;Mohri 2017;Kouadio et al 2022). …”