Machine Learning-based systems act on flow features derived through exporting flow procedures. The notable emergence of Machine Learning and Deep Learning (DL) based reports presuppose that the flow of information, such as the average packet capacity, is gleaned from every packet. On common devices, However, when packet sampling is unavoidable, flow exporters are frequently used in practice. Since the flow of information is derived from a sampled group of the packets rather than the entire traffic stream, the usefulness of Machine Learning-based results with the use and existence of such samplings is still up for debate. In this study, we are going to investigate in what ways the effectiveness and performance of these ML-based are affected by packet sampling. Our suggested evaluation method is resistant to various flow export stage settings, in contrast to earlier studies. Hence, it can provide a robust evaluation even in the presence of sampling.