We have a pretty simple R forecasting problem, using randomForest. Our
script is below. We have data with about 6 columns, and 625 rows to do the
training on, and then we're trying to forecast 90 or so rows.
The x matrix has 3 years worth of sales data, starting at 6:15am. However,
for some
Well, if you look at your data more carefully, you will see that the
histogram of y is heavily skewed towards 1 (small values). The 91/625
quantile is still 1 (there are 192 1s). It is therefore not surprising
that RF comes up with mostly 1s (in my attempt it came up with two 2s,
but that is a bit
Peter,
Thanks for checking on this! We're having some difficulties forecasting
sales ... randomForest is doing a great job on days and periods where there
is high variability. However, such as in this case, we do get
dates/times/departments where it simply forecasts the most frequent value
(I
3 matches
Mail list logo
