Forecasting call volumes over short intervals using R

I am trying to do a basic forecast of call volumes using the forecast library for R. I am not having too much trouble forecasting on a daily or monthly interval, however when I try to forecast on an hourly or or quarter hourly basis I am having problems. The variation between points is so small as to be non existent.

The data given are made up of aggregates for each period going back up to 24 months (I've tried shorter periods with no success). The points are only between 8am and 8pm, meaning that there are 12 points per day for hourly data and 48 points per day for quarter hourly.

I believe the issue lies either with cycle_length or possibly needing to try a different smoothing algorithm, unfortunately my experience with statistics is so small as to be negligible.

My code is as follows:

data <- read.csv(input_csv, header=FALSE)

ts_data <- ts(data, frequency = cycle_length)

ts_forecast <- forecast(ts_data, h = forecast_periods)

write.csv(ts_forecast, output_csv, row.names=FALSE)


Edit: As requested, a chart of some of the data. The full data set is 24 months long (I've experimented with using all or part of it). Obviously this is impractical to put up, so here is a randomly selected 7 day snapshot. Quarter hourly intervals 12 hours per day, meaning 48 points per day.

• Hi Akrist. Could you put up a plot of the call series at quarter hourly intervals so that we can see what it looks like, and what might be appropriate? – Deathkill14 Apr 29 '14 at 8:51
• I've added a chart to the original post. – akrist Apr 30 '14 at 1:28
• You can post the data to dropbox.com and then share the link here so we can take a look. State the beginning date of the data and the country of origin. – Tom Reilly Apr 30 '14 at 19:40