Anybody here familiar with detrended fluctuation analysis (DFA)?

To those who are familiar with detrended fluctuation,

Answer 1

[Ayush Modi]

Hi Hunter,

As per my understanding, you are trying to perform a Detrended Fluctuation Analysis on a sample data but getting negative or very small value. The attached script, when executed is throwing an error saying “Invalid data type. First input argument must be numeric or logical”.

Assuming first argument “file_name” is a string, it is assigned to variable x which is used as an input to “cumsum” function. “cumsum” function requires vector or matrix as an input.

Here is an example showing how you can perform the detrended Fluctuation Analysis:

 x = load(file_name);
   
% Number of points in the time series
numberpoints = length(x);

% Define the minimum and maximum box sizes
MinBox = 4;
MaxBox = floor(0.25 * numberpoints);
BoxSizeDensity = 4;
LogScaleFactor = 2^(1/BoxSizeDensity);

% Initialize arrays to store the results
log_points_in_box = [];
log_Q = [];

% Perform DFA
BoxSize = MinBox;
while BoxSize <= MaxBox
    % Initialize fluctuation for this box size
    F = 0;
    
    % Number of boxes at this box size
    numBoxes = floor(numberpoints / BoxSize);
    
    for i = 1:numBoxes
        % Indices for the data in this box
        idxStart = (i - 1) * BoxSize + 1;
        idxEnd = i * BoxSize;
        
        % Data in the box
        boxData = x(idxStart:idxEnd);
        
        % Linear detrend the box data
        p = polyfit((1:BoxSize)', boxData, 1);
        boxDataDetrended = boxData - polyval(p, (1:BoxSize)');
        
        % Calculate fluctuation for this box
        F = F + sqrt(mean(boxDataDetrended .^ 2));
    end
    
    % Average fluctuation over all boxes
    F = F / numBoxes;
    
    % Store the results
    log_points_in_box = [log_points_in_box; log10(BoxSize)];
    log_Q = [log_Q; log10(F)];
    
    % Update the box size for the next iteration
    BoxSize = BoxSize * LogScaleFactor;
    BoxSize = round(BoxSize);
end

% Perform linear regression on log-log data
coeffs = polyfit(log_points_in_box, log_Q, 1);
H = coeffs(1); % Slope of the line is the Hurst exponent
r_line = polyval(coeffs, log_points_in_box);
r2 = corrcoef(log_points_in_box, log_Q);
r2 = r2(1, 2)^2; % Coefficient of determination

Please refer to the following MathWorks documentation for more information on “cumsum” function:

https://www.mathworks.com/help/matlab/ref/cumsum.html#btrgrnv-1-A

I hope this resolves the issue you were facing.