1
$\begingroup$

I got some users' history data and generated some sequences of real numbers. The length of each sequence is between 15 and 25. What's more, I do not know whether these sequences have patterns and the frequency is not known as well.

My goal is using each sequence to predict its next value, and then I use auto.arima in R to do this. However, the accuracy of the prediction is low.

Anyone have any good ideas to improve the accuracy?

One of these sequences is:

    1.5959709882736206  
    0.7300914525985718  
    2.0011744499206543  
    3.6755871772766113  
    0.8066112399101257  
    1.3413848876953125  
    3.371157646179199  
    0.4400146007537842  
    2.637667655944824  
    2.1453769207000732  
    2.341433048248291  
    2.3429665565490723  
    1.1187453269958496  
    1.4169363975524902  
    3.328829050064087  
    4.157748699188232  
    3.9255290031433105  
    2.7843635082244873 
$\endgroup$

1 Answer 1

1
$\begingroup$

Chelsea,

You have constricted your answer to a sample space that is too small. You need to be considering deterministic variables instead of relying upon stochastic only. This model has two trend variables. One beginning at period 1 and another at 9. The 4th observation is an outlier.

Y(T) = 1.2875
+[X1(T)][(+ .0461)] :TIME TREND 1 +[X2(T)][(+ .126)] :TIME TREND 9 +[X3(T)][(+ 2.2038)] :PULSE 4 + + [A(T)] enter image description here

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.