The identification of cardiac anomalies in fetus during the pregnancy can be find by the ECG monitoring system non-invasively, which provides visual information about the mother’s and fetus’s heartbeats. The analysis of graphical pattern of fetus heartbeats by different machine learning methods can detect and classify different abnormalities in heart during the developing phase in 4 to 6 gestation period of mother. In this paper, we provide a systematic survey on fetal arrhythmia detection by reviewing the recent papers on accuracy and machine learning models to summarize best machine learning models and find the gaps to improve the accuracy. The majority of the papers uses CNN or LSTM and trained their model on dataset provided by physionet (NIFEDBA) and MIT-BIH with the best accuracy of 98.56%. This paper find the gaps in signal-to-noise ratio (SNR) and overlapping ECG of mother and fetus as a result impacting factor which is not appropriately considered in most of the papers. The multilevel classification of tachycardia and bradycardia is also the potential deficiency which could improve the classification of arrhythmia.