The rise of machine learning has seen an increase in digital methods for use within health care. Arrythmia detection is one of the areas where this increase is obvious. However, many machine learning methods for arrythmia detection utilise models that are computationally expensive, such as convolutional neural networks, CNNs. This thesis examines whether it is viable to use the Fast Fourier Transform to transform an Electrocardiogram, ECG, signal into its frequency components before training a neural network, NN, on the data. This could allow for a lower computational cost and wider availability of arrythmia detection technology. The results from the model were compared to that of a CNN trained on time domain data. The results show that the CCN model outperforms the NN trained on FFT transformed data but the performance of the model still indicates that valuable information about heart arrythmias does exist within the frequency space. This suggests a potential for future work on the subject.