Conjunct applicability of MCDM-based machine learning algorithms in mapping the sediment formation potential

This study evaluates the applicability of multicriteria decision-making (MCDM) methods, including SAW, VIKOR, TOPSIS, and Condorcet algorithm based on game theory and machine learning algorithms (MLAs) including K-nearest neighbor, Na & iuml;ve Bayes, Random Forest (RF), simple linear regression and support vector machine in spatial mapping of sediment formation potential in Talar watershed, Iran. In the first approach, MCDM was used, including SAW, VIKOR, TOPSIS, and Condorcet's algorithm based on game theory. To this end, a decision matrix for MCDM was first created based on the factors affecting sediment formation potential. In the next step, various MLAs were used to construct a distribution map of sediment formation potential. Finally, a distribution map of sediment formation potential was constructed in very low to very high classes. The summary of the results of prioritizing sub-basins based on sediment formation potential using multi-criteria decision-making methods showed that sub-basin SW12 had the highest sediment formation potential based on VIKOR, TOPSIS, and Condorcet methods. The results of sediment formation potential modeling using different machine learning algorithms showed that based on the values of error statistics, the algorithm RF with the values MAE = 0.032, MSE = 0.024, RMSE = 0.155, and AUC = 0.930 was selected as the most optimal algorithm. On the other side, the correlation matrix and Taylor diagram (Figs. 10 and 11) also showed that RF algorithm modeling with the slope factor had the highest correlation with a value of 0.84. Also, the LS factor with a correlation coefficient of 0.65 after slope had the highest correlation with the RF model in sediment formation modeling. The sediment formation potential map based on the RF algorithm shows that the amount of sediment increases from the downstream to the upstream side of the Talar watershed.