Using algorithmic game theory to improve supervised machine learning: A novel applicability approach in flood susceptibility mapping

• Published in: Environmental science and pollution research international Vol. 31; no. 40; pp. 52740 - 52757
• Main Authors: Nasiri Khiavi, Ali, Vafakhah, Mehdi
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2024; Springer Nature B.V
• Subjects: Algorithms; Aquatic Pollution; Atmospheric Protection/Air Quality Control/Air Pollution; Classification; Curvature; Earth and Environmental Science; Ecotoxicology; Environment; Environmental Chemistry; Environmental Health; Euclidean geometry; Flood mapping; Floods; forests; Game Theory; Iran; Learning algorithms; Machine Learning; Mapping; prioritization; Rangelands; regression analysis; Research Article; Similarity; streams; subwatersheds; Supervised learning; Supervised Machine Learning; Support Vector Machine; Support vector machines; Topography; Waste Water Technology; Water Management; Water Pollution Control; Watersheds; Iran; Artificial Intelligence (AI); Geospatial modeling; Python programming language; Multi-Criteria Decision Making (MCDM); Flood modeling
• ISSN: 1614-7499; 0944-1344; 1614-7499
• DOI: 10.1007/s11356-024-34691-y

This study was carried out with the aim of applying Condorcet and Borda scoring algorithms based on Game Theory (GT) to determine flood points and Flood Susceptibility Mapping (FSM) based on Machine Learning Algorithms (MLA) including Random Forest (RF), Support Vector Regression (SVR), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) in the Cheshmeh-Kileh watershed, Iran. Therefore, first, FS conditioning factors including Aspect (A s ), Elevation (E l ), Euclidean distance (Euc), Forest (F), NDVI, Precipitation (P), Plan Curvature (P l C), Profile Curvature (P r C), Residential (R e ), Rangeland (R l ), Slope (S l ), Stream Power Index (SPI), Topographic Position Index (TPI), and Topographic Wetness Index (TWI) were quantified in each Sub-Watershed (SW). Based on this, flood and non-flood points were identified based on both GT algorithms. In the following, MLAs including Random Forest (RF), Support Vector Regression (SVR), Support Vector Machines (SVM), and K-Nearest Neighbors (KNN) were used for the distributional mapping of FS. Finally, based on optimal conjunct approaches, FS maps were presented in the study watershed. Based on the results, among the conjunct algorithms in FS classification, RF-Condorcet and RF-Borda models were selected as the most optimal MLA-GT hybrid models. The upstream SWs were highly susceptible. Also, the effectiveness of NDVI and forest conditioning factors in each classification approach was high. The similarity of SW prioritization based on Condorcet algorithm with RF-Condorcet algorithm was about 86.70%. Meanwhile, the degree of similarity in RF-Borda conjunct algorithm was around 73.33%. These results showed that Condorcet algorithm had an optimal classification compared to Borda scoring algorithm.