Distributed energy resources (DER) systems introduce uncertainties in the electrical grid that cannot be addressed by classical deterministic methods. Power system analytic tools, such as Load Flow (LF), should be revisited to address such uncertainties. Probabilistic Load Flow (PLF) provides a solution to this problem by handling these uncertainties as random variables. Among the existing sampling methods, the Unscented Transform (UT) has provided reliable and fast estimations to highly correlated systems. However, due to the large-scale integration of DER, faster and more accurate estimations are needed. In this paper, four variants of PLF, based on the UT method, with the effects of their weighting parameters are described, analyzed and compared in the IEEE 30 and IEEE 118 test cases. Then, based on this analysis recommendations and guidelines are proposed to improve the estimation based on the characteristics of the system and UT PLF method. Results are compared with previous literature benchmarks and Monte Carlo Simulation (MCS). The proposed recommendations significantly improved accuracy and required less computational time.