Fit exponential curve through data points in Matlab

Question

Having data of an exponential decay available, I would like to fit a curve through it. How can I do that in Matlab?

There is a nice demo on the Statistics Toolbox product page that discusses problems that can occur if you transform a nonlinear model to a linear one. http://www.mathworks.com/products/statistics/demos.html?file=/products/demos/shipping/stats/xform2lineardemo.html The preferred method is to use nonlinear regression; deriving a set a "smart" starting conditions using the "linearize, least squares, delinearize" technique. — richard willey, Mar 24 '11 at 17:40
The comment by Richard Willey should really be the (accepted) answer. Linearizing and applying least squares as recommended in some of the answers is not a good idea since the transformation will give too much weight to small values. Better use nlinfit. — Lukas, Mar 26 '12 at 12:54
Link from @richardwilley has been broken by The MathWorks. Updated version appears to be [here](http://www.mathworks.com/help/stats/examples/curve-fitting-and-distribution-fitting.html). — horchler, May 15 '15 at 20:41

score 5 · Accepted Answer · edited Jun 24 '13 at 03:20

5

Try this:

ft=fittype('exp1');
cf=fit(time,data,ft)

This is when time and data are your data vectors; time is the independent variable and data is the dependent variable.

This will give you the coefficients of the exponential decay curve.

edited Jun 24 '13 at 03:20

eggy

2,836
3
23
37

answered Jun 24 '13 at 02:50

Adiel

3,071
15
21

how would one plot this? – Nick Dec 11 '14 at 11:53
@Adiel When i tried this I get an error as `Undefined function 'diff' for input arguments of type 'cfit'. Error in cfit/plot (line 64) if any(diff(xdata)<0) Error in slopeFirst2hours (line 6) plot(time, cf, 'r') ` Can you tell me what I am doing wrong – clarkson Sep 05 '17 at 11:23
@clarkson You are right. It should be plotted with a single command- `plot(cf,time,data)` – Adiel Sep 05 '17 at 12:06

score 2 · Answer 2 · answered Mar 22 '11 at 14:48

2

If by fit you mean least squares, you should try lsqcurvefit

answered Mar 22 '11 at 14:48

user389419

243
1
2
8

score 1 · Answer 3 · answered Mar 22 '11 at 17:33

1

cftool(X,Y) is the way to go. here's some linkage:

LINK1 LINK2

answered Mar 22 '11 at 17:33

Marcin

3,437
1
22
15

score 0 · Answer 4 · edited Jan 06 '14 at 11:36

0

Matlab has a function called polyfit. It can fit curve to a data which can be represented in the form a*X^n+b*X^(n-1)+.....z. However if you are sure that the data is of some exponential decay you can try taking logarithm of the data first and then using the polyfit function. I thing that will work.

edited Jan 06 '14 at 11:36

BenMorel

34,448
50
182
322

answered Mar 22 '11 at 14:56

bubble

3,408
5
29
51

score 0 · Answer 5 · answered Mar 22 '11 at 15:00

0

Linearise, least squares, delinearise :-)

answered Mar 22 '11 at 15:00

regularfry

3,248
2
21
27

You are right, although I thought there might be a package that does that for me automatically it is probably less time consuming to just do it manually rather than searching for it :) – Ingo Mar 23 '11 at 10:10
This is pretty short for an answer, but I upvoted it from -1 because it is helpful for users without the Curve Fitting toolbox. – FvD Jan 06 '15 at 08:11

Fit exponential curve through data points in Matlab

5 Answers5

Linked