The piezocone penetration test (CPTu) is one of the primary screening tools used by the mining industry to evaluate whether tailings are susceptible to liquefaction (static or cyclic). Liquefaction analysis using the CPTu is influenced by the fines content of the soil, and most CPT screening tools empirically use the behaviour type index (Ic) to correct for fines. The issue with this approach is that such empirical relationships have been developed using data collected from natural soils and their applicability to different tailings deposits needs to be studied. In this paper, we explore the use of machine learning to develop a site-specific model to estimate fines content of tailings using gamma piezocone penetration testing (GCPTu) measurements. The results are then compared to the fines content derived from the soil behavior type index (Ic) proposed by Robertson and Wride (1998). A dataset of paired GCPTu and laboratory results were compiled and a support vector machine (SVM) algorithm was employed to calibrate a model to estimate tailings fines content from in-situ GCPTu data. We calibrated GCPTu data to geotechnical 75 μm fines content measured in the laboratory. The geotechnical fines content is the fines content as a percentage of solids. The development dataset was comprised of paired GCPTu-laboratory results collected in 2009-2019 from a mining area with predominately clay mineral and quartz sand tailings. Results show that the developed model predicts the geotechnical 75 μm fines content of the studied deposits with 7.8% error. The Ic based method estimates the fines content with 31% error. Evidently, the Ic method is not applicable to the studied tailings deposits. It is appreciated that many tailings fines do not contain clay, and that the success on this program may not be repeatable everywhere. However, the authors wish to point out that the application of site-specific machine learning algorithms may be used to significantly improve empirical geotechnical parameters.