Fetal electrocardiography (FECG) is used in the process of prenatal monitoring because it is a direct diagnostic technique that helps to assess cardiac dysfunction and to diagnose fetal arrhythmias. However, expansion, selection, and improvement of FECG signals are still problematic due to the presence of overlapping MECG, noise, and fetal signal fluctuations. Old school signal processing techniques including filtering and blind source separation have been used but draw the line at handling complex and noisy environments adequately. Current progress in the direction of ML as well as DL holds promising capabilities to integrate feature extraction, noise removal, and real-time computation. In this paper, the current FECG extraction methods have been examined with the corresponding traditional and new conventional with the help of an ML approach. These include the delineation of problems and successes, with a crucial focus on what the clinical application of ML means to prenatal diagnostics. The survey also reveals the limitations of prior and ongoing research and highlights directions for improvement and further developments in FECG analysis.