A Fast Fourier Transform Method for Measuring Randomness Level of a Time Series Dataset

Time series prediction is applied in various fields and is performed in a time-spatial dataset. To be able to be predicted, a dataset must have adequate relation between data in the time interval. So far, the method to examine it is by using correlation, which is widely used for measuring similarity between variables. In the time series data, the correlation value is autocorrelation, which defines the correlation between an actual value and time-lag values in the future sample. Furthermore, for highly random data, the autocorrelation value drops significantly after even a one-lag iteration. Then, it only resembles a Gaussian or white noise that is difficult to distinguish. Therefore, this study extends the calculation of the correlation by proposing a fast Fourier transform (FFT). The result of the proposed method shows that the randomness of various datasets could be better distinguished.