pso-optimized decision tree and svm for fluid saturation prediction

This study presents an ml framework to predict sw by employing two algorithms—dt and svm—optimized using the pso technique. input data are composed of the lithology logs (sp and gr), resistivity logs (shallow, rxo, and deep, rd), compensated neutron log (cnl), and depth. before modeling, data outliers were detected and removed using a gaussian elimination method. results indicate that the svm model is more accurate, achieving a coefficient of determination (r2) of 0.947 on test data and 0.989 on training data. the study demonstrates that integrating pso-based optimization with systematic data preprocessing significantly enhances the accuracy of ml models in sw estimation, offering an efficient and cost-effective alternative to conventional interpretation methods.