TIME SERIES VIA

FRACTALS
GEOMETRY
K.BHAVANI(MA14M003)
R.HEMA(MA14M012)

Why?
Where?
How?

Why time series via fractals?

Chaotic systems exhibit rich and surprising mathematical structures.

Deterministic chaos provides a striking interpretation for irregular

temporal behaviour.

The most direct link between chaos theory and the real world is the
analysis of time series of real systems in terms of nonlinear
dynamics. Sometimes time series are so complex that they could not
be examined by traditional methods. In such cases time series data is
analysed via fractals.

Where?

Geo physics time series(recording of temperature, precipitation)

Medical and physiological time series(recording of heartbeat, ECG,

blood pressure, respiration, blood flow, glucose levels.. etc.)

Astrophysical time series

Technical time series(Internet traffic, highway traffic )

Finance and economy.

Application of fractal dimension in ECG data:

ECG curve

Above figure defines the shape and size of the P-QRS-T wave of the
electrocardiogram(ECG), this is essential to determine the classification
of heart failure from ECG. The characteristics of all interval (PR, QRS,
QT, RR) can change by specific heart failures. ECG signals are widely
examined signals, thinking the significance in humans life.

Decisions about ECG signals were made through handling the intervals
between P-QRS-T waves..

How

Three kinds of fractal dimension methods in fractal geometry are used.

1.Correlation method
2.Hausdorff method
3.Box counting method

Fractal dimension:

It is a measure of thespace-fillingcapacity of a pattern that

tells how a fractal scales differently from thespaceit is
embedded in.

a fractal dimension does not have to be an integer.

Correlation method:

Autocorrelation:

Autocorrelation function is a time dependent function.

the autocorrelation function measures the average magnitude of the

correlation of two points separated in time with a lag t.

When there is no correlation in the time series then the expectation value
of correlation function is zero. Autocorrelation function is computed by
the formula.
A()=-)(-)/-)^2

represents ith element of the series, is time delay (Optimal time lag)
and N is equal to length of the series.

is determined from the shape of the correlation function, as the

position of its first decrease under 1/e or 1/10, and this value is used to
find the embedded dimension in phase space, which gives the
correlation dimension value.

Functions related to autocorrelation are directly the main way in order

to estimate the fractal dimension by means of power law relations.

Quantities varying with time often turn out to have fractal graphs. One
way in which their fractal nature is often manifested is by a power-law
behaviour of correlation between measurements separated by time h.

A measure of the correlation between f at times separated by h is

provided by the autocorrelation function

C(h) = ) )dt

Autocorrelations provide several methods of estimating the dimension

of the graph of a function or signal f .

The autocorrelation function C(h) or equivalently, the mean-square

change in signal in the time h over a long period could be computed,
and so
2[C(0)-C(h)] dt

If the power law behaviour is

C(0)-C(h) c

Then s is said to be dimension.

Hausdorff method:
The

direct calculation of the Hausdorff dimension is often intractable.

It is calculated by using Hurst exponent method.

Hurst exponent is calculated by using Rescaled Range analysis.

To do this, a time series of full length is divided into a number of

shorter time series and the rescaled range is calculated for each of
the smaller time series. A minimum length of eight is usually chosen
for the length of the smallest time series. So, for example, if a time
series has 128 observations it is divided into:

two chunks of 64 observations each

four chunks of 32 observations each

eight chunks of 16 observations each

16 chunks of eight observations each

Steps for estimating the Hurst exponent after breaking the time series into
chunks:

For each chunk of observations, compute:

The mean of the time series,

A mean centered series by subtracting the mean from the series,

The cumulative deviation of the series from the mean by summing up the mean
centered values,

The Range (R), which is the difference between the maximum value of the
cumulative deviation and the minimum value of the cumulative deviation,

The standard deviation (S) of the mean centered values

The rescaled range by dividing the Range by the standard deviation.

Finally, average the rescaled range over all the chunks. i.e. H

Hausdorff dimension has the advantage of being

defined for any set, and is mathematically convenient,
it is based on measures, which are relatively easy to
manipulate.
The Hausdorff dimension calculate by using the formulae

DH= -H
where H- hurst exponent(0<H<1)

H can be calculated by the following formulae

where n- no.of data points in time series

R(n)-range value
S(n)-standard deviation
c-constant

Box counting method:

One of the simplest and most intuitive method of fractal dimension is

box counting with dimension .

Given a fractal set in a d-dimensional Euclidean space, gives the

scaling between the number of d-dimensional boxes needed to cover
the set completely, and the boxes size .

For the ECG time series graph given above the different size of the boxes
the no. of boxes tabulated below

data given above in the table will able to plot a regression line.
The

Taking

the general fractal dimension formula, the logarithmic N(E) values

and the logarithmic E values have a linear relationship.

=
There

is a facility to find the optimal box number to calculate the boxcounting dimension.

simple linear regression, Least Square Method, which are an useful and
popular technique in the field of statistics, to find the box counting
dimension via the slope of the regression line.

Fitting Regression line:

The method of least squares is a standard approach to the

approximate solution of overdetermined systems, i.e., sets of
equations in which there are more equations than unknowns. "Least
squares" means that the overall solution minimizes the sum of the
squares of the errors made in the results of every single equation.

The least squares line uses a straight line y=a+bx

To approximate the given set of data, (,),, (,),where n2.

The best fitting curve f(x) has the least square error, .i.e.

where a and b are unknown coefficients while all and are given. To
obtain the least square error, the unknown coefficients a and b must yield
zero first derivatives.

By expanding above equations we will get

where b gives the slope of the regression line.

In the lightening of this expression, the box-counting dimension value can
be obviously seen from the regression equation, because it is a linear
equation and the value of b gives the slope of the regression line, exactly
being equal to the value of box-counting dimension.

In the below chart 1.4068 is the box dimension of the data.

Chart Title
10

Linear ()

9 f(x) = - 1.41x + 9.43

R = 1
8
7
6
Ln of 5number of boxes
4
3
2
1
0
0

3
Ln of box size

Accuracy of Fractal Dimension Methods

We need to prove the accuracy of these 3 fractal dimension methods.

Therefore a sinusoidal wave was examined using correlation, Hausdorff

and box-counting dimensions, in the lightening of the knowledge that
this sinusoidal wave should have the value of nearly one as dimension.

This plot shows the correlation dimension of the above sinusoidal

wave.

The Hurst exponent and Hausdorff Dimension Graph of Sinusoidal Wave

is

Hurst exponent value is calculated from the slope of the line.

Hurst exponent value=0.991829, Hausdorff dim=1.008171.

Box dimension:
SLOPE=1.8314
R

N(r)

234550

59010 14805

16

32

3763 972 252

64

128

256 512 1024

63

20

Chart Title
14
12

f(x) = - 1.83x + 12.09

10 R = 1
8
LN(N(R))

6
4
2
0
0

4
LN(R)
Linear ()

CONCLUSION:
FRACTAL DIMENSION TYPE

DIMENSION VALUE

CORRELATION

1.002

HAUSDORFF

1.008171

BOX

1.8314

SINUSOIDAL WAVE

According to these results, two techniques indicate high degree of accuracy

apart from Box dimension, because the value of sinusoidal wave dimension
should be nearly 1, Box dimension value is quite high.

Applications:

By using 3 fractal dimensions ECG data is analysed. It contains 2700

data points.

Plot of the time series data:

Correlation dimension:

Embedded points = 2692

Time lag =6

Oribital lag =56

Correlation dimension
1.924 where embedded
Dimension is 2.

Hausdorff dimension
H value is 0.995760

Hausdorff dim=2-H=1.004240.

Box dimension:

THANK YOU