The term fetal arrhythmia refers to irregular fetal heart rhythms, with the established heart rate range being 110 to 160 beats per minute (bpm). Fetal arrhythmias occurin 1 to 2% of pregnancies; and although the majority of these are benign and transient, both tachyarrhythmia and bradyarrhythmia in some cases can indicate a serious condition for the fetus or the mother. Thus, a persistent fetal arrhythmia can lead to decreased cardiac output, heart failure, hydrops, and even fetal demise. As a result of this situation, an early diagnosis is crucial to adequately address this condition and reduce related mortality. Therefore, this study proposes the use of SVD entropy for characterizing ECG data from 6 channels (fetal and maternal), aiming to differentiate between healthy and diseased individuals. Consequently, a neural network could classify them, thus enabling a non-invasive early diagnosis of fetal arrhythmia. Additionally, it aims to enhance the performance of this technique by employing genetic algorithms for data augmentation and selecting the optimal architecture for the neural network, thereby ensuring a global accuracy of over 88% in fetal arrhythmia risk stratification.