python - dynamic forecast adjustment on error testing

Question

I have multiple models running to forecast healthcare usage and have the daily actual number feeding in to my database.

I want to be able to optimize the forecasted view taking into account the errors that the other models had with the aim of removing the worst performing model over the last 4 days to build an optimized new actual forecast.

I tried to add back the delta created by the worst performing model but that didn't work really well.

Is there any statistical/mathematical approach that can be done to solve this?

sample df- f being the forecast and M1,M2,M3 being the models, with the last 2 columns being the output dataframe

Date        f   A   M1  M2  M3  Adjusted fc(last 3 days)  Adjusted fc(last 5 days)    Adj best guess
01-01-2016  9   17  14  2   10      
02-01-2016  13  1   15  5   18      
03-01-2016  6   2   3   1   13      
04-01-2016  3   8   6   3   1       
05-01-2016  8   15  15  0   8       
06-01-2016  8   12  6   5   12      
07-01-2016  11  18  15  17  2       
08-01-2016  12  3   12  17  7       
09-01-2016  3   14  2   3   4       
10-01-2016  8   12  16  5   2       
11-01-2016  13  14  15  8   16      
12-01-2016  11  11  10  18  6       
13-01-2016  8   1   1   7   17      
14-01-2016  11  3   4   19  11      
15-01-2016  6   2   11  3   4       
16-01-2016  9   12  8   8   12      
17-01-2016  10  7   10  6   13      
18-01-2016  13  19  9   19  11      
19-01-2016  7       2   10  10      
20-01-2016  11      18  3   11      
21-01-2016  15      11  19  16      
22-01-2016  13      15  14  9   

Answer 1

等待大神解答