2

I have to do some experiment and measure it on a specific time 0, 3, 6, 9, 12, 18, 24, 36 months $$ \begin{array}{l|c|c|c|c|c|c} \text{Month} &ID1 & ID2 & ID3 & ID4 & ID5 & ID6\\\hline 0 & 101.7& 102.6& 101.7 &100.5 &100.4 &103.4\\ 3 &103.4 &103.3 &101.4 &101.7& 100.5 &101.2\\ 6 &100.7 &103.7 &101.6 &102 & 102.9& 102\\ 9 &100.2 &100.6& 101.2& 97.9& 98.7 &99.5\\ 12 & 99.8 &100.1& x & 98.6& 100.4 &100.1\\ 18 &98.7& x& x & x & x &x\\ 24 &101.3 &x &x & x & x& x\\ 36 &100.8 &x &x& x &x & x\\ \end{array} $$ Basically, all the numbers are from the measurement. So, could anyone please suggest how do I predict the numbers represented as "x" here ? Assume each ID are different type of items/products/equipment to be measured.

I would prefer to use the arithmetic mean and variance for the prediction. However, if anyone could suggest better idea, it would be nice.

Thank you in advance, PTP

Arthur
  • 199,419
Modnoiz Antonio
  • 23
  • 1
  • 3
  • I made sure your table was readable (the text editor doesn't interpret newlines, spaces and tabulators as you might expect it to, so your numbers were a mess). Could you go through the table and make sure I didn't ruin anything? – Arthur Aug 26 '14 at 11:01
  • Hello Arthur, Thank you for your help. :) – Modnoiz Antonio Aug 27 '14 at 11:56

1 Answers1

1

It appears your data have some suggestion of trend (negative). I'd suggest using Double Exponential Smoothing, as opposed to the sample mean, which could end up overpredicting your future values.

I applied double exponential smoothing to your data, and tentatively found that $\alpha=0.6,\gamma=0.1$ to be reasonably good. See below.

enter image description here

  • Hi Eupraxis, I am not familiar with the double exponential. So, my question is, this method can apply to the other data sets? – Modnoiz Antonio Aug 27 '14 at 11:57
  • @ModnoizAntonio yes, it is a general method for generating forecast values. –  Aug 27 '14 at 12:21
  • @eupraxis1991 Is it working for both positive and negative trends right? If yes, then thank you for your suggestions. :) – Modnoiz Antonio Aug 27 '14 at 14:49
  • @ModnoizAntonio yes, it will adapt to changes in trend, it only requires initial estimates of the mean value and trend for a given time series, then it updates itself after new data come in. –  Aug 27 '14 at 15:07