I am currently working on a regression problem which requires me to predict the costs of a fixed asset. I have used several variables to do so and derived a predicted cost. However, my superior has wanted me to incorporate time as a variable in the regression model which I am at a loss on how to do so. My data set looks something like this which does not resemble a time series dataset.

Name    Capacity    OEM      Country   Date of valuation  MONTH   YEAR   Cost
A1       220        JAPAN    JAPAN     1/1/2012            1      2012   300,000,000 
A2       220        JAPAN    JAPAN     1/1/2012            1      2012   300000000
B1       400        CHINA    CHINA     1/3/2013            3      2013   475000000
B2       400        CHINA    CHINA     1/3/2013            3      2013   475000000   
B3       400        CHINA    CHINA     1/3/2013            3      2013   475000000
B4       400        CHINA    CHINA     1/3/2013            3      2013   475000000
C1       750        INDIA    USA       1/5/2016            5      2016   268000000
C2       750        INDIA    USA       1/5/2016            5      2016   268000000  

The variables that I have used are capacity, OEM and country. Any help on how to incorporate time to my regression problem is welcomed.


Theres an approach I'd take which consists of two steps. The second is optional buy highly recommended.

  1. create variables for day, month, year separately.
  2. process those numbers as cyclic features (day 31 is as close to day 1 as to day 30, for example. If you don't do anything about that, and store days from 1 to 31, there's no place considering the cyclic nature of the feature). This post exaplins this in detail.

Hope this helps.

EDIT1: Also, not asked, but depending on the algorithm you're using, you may want to normalize the numerical features you have.

  • $\begingroup$ Thanks for your help. I have actually think of this method in dealing with time variables. Can I ask if the time variables must be a continuous variable? Also, my superior is very insistent for the time variables to follow PPI indices which I do not think is correct(E.g PPI indices in the base year 2012=100 will be allocated to 1, PPI indices in year 2017=120 will be allocated to 1.2). How can I resolve this issue? $\endgroup$ Jul 4 '19 at 1:18
  • $\begingroup$ On trigonometric functions, it should be continuous type. On date decomposition I don't think that would be correct. Day number is either 3 OR 4, for example, day 3.5 makes no sense, unless you consider it to be day 3 at 12.... But you should process those variables accordingly; a priori, with what we've talked till now, nothing indicated a possibility to use continuous values. $\endgroup$
    – 89f3a1c
    Jul 4 '19 at 15:55
  • $\begingroup$ As of PPI indices: I don't really know anything about them; if you could point out a source where I could read something about it, I'd be happy to think further. $\endgroup$
    – 89f3a1c
    Jul 4 '19 at 15:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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