A statistical method used to quantify the fractal scaling properties and long-range correlations in a signal.DFA is widely used in fields ranging from physiology to climatology to detect hidden patterns in complex systems. For example, it can distinguish between a healthy heart's beat-to-beat variability, which shows fractal-like scaling, and an unhealthy heart that may exhibit more random behavior. In physics, it helps analyze signals from turbulence, earthquakes, and even DNA sequences, where long-range correlations indicate underlying structure.
DFA was introduced in 1994 by physicists C.-K. Peng, Shlomo Havlin, H. Eugene Stanley, and colleagues at Boston University, originally to analyze DNA sequences. It was later adapted for heart rate variability and has since become a standard tool in many fields.
The method works by first integrating the time series to convert it into a random walk-like path. This integrated signal is then divided into windows of varying lengths. In each window, a local trend (usually a straight line) is fitted and subtracted, removing any slow drift. The average fluctuation around the trend is computed for each window size. If the fluctuation grows as a power of the window size, the exponent indicates the nature of correlations: an exponent of 0.5 suggests white noise (no memory), while exponents between 0.5 and 1 indicate long-range positive correlations, and exponents less than 0.5 indicate anti-correlations.
DFA is used in medicine to assess cardiac health by analyzing heart rate variability, in finance to study market volatility, and in climate science to detect long-term trends in temperature and precipitation records.
DFA was originally developed to find long-range correlations in DNA sequences, where it revealed that the arrangement of nucleotides is not random but follows fractal patterns, hinting at evolutionary constraints.






