Abstarct: Transductive support vector machine spectral clustering (TSVMSC) inference algorithm considers spatial heterogeneity in all models, and is flexible and can be used with any type of data to identify the dominant patterns and structures in the multi-variable data. This algorithm has a stable performance and a high level of accuracy. The purpose of this thesis is to improve the integration of the results of inversion of refraction seismic tomography and resistivity data using the TSVMSC algorithm as a knowledge-baxsed approach. For this purpose, the seismic refraction tomography and electrical resistivity data acqired from one of dams in Iran have been analyzed. To verify the modified model obtained as a result of integration of the data by the TSVMSC algorithm in a controlled implementation, we have used a synthetic example of a heterogeneous unconsolidated aquifer. After the inversion of synthetic and field data, for the integration of the results, k-means, FCM, Gustaffson-Kessel and TSVMSC algorithm have been used. baxsed on the results, the separation of alluvial regions, submerged bedrock and high quality bedrock has been made, and in this way, the results have been validated. Moreover, the results have shown that the TSVMSC algorithm has been more accurate and more effective in integrating heterogeneity and complexity of the two models. The Gustaffson-Kessel fuzzy cluster has also been more successful than other clustering methods used in this research. baxsed on the validation of the seismic refraction tomography data inversion results, the estimation error was calculated to be 30.95%. baxsed on the results of experiments, the experimental relationships for compressive and shear velocity were obtained, and consequently, correlation coefficients of 76.5 and 77.5 percent were obtained for compressive velocity and shear velocity with porosity, respectively.