Time Series Forecasting on minute transaction data

[anilaaryan]

Dear Team,

I am new to time series. I have a problem in hand where i am trying to forecast transaction values for the period of 30 days. For this I have a historical data for the last 10 years with transactional data recorded every 15 mins. I have used prophet and forecasted for 30 days. But my output looks little strange. Below is the output for your reference. It is just using the last record and with the same time period it is forecasting for next 30 days i.,e at 19:00 hours for next 30 days. How Can i forecast every 15 mins for next 30 days? Please advise on where I am going wrong.

I have used the below code to do that <sales_data[['Date']] <- as.POSIXct(transact_data[['Date']], format = "%Y-%m-%d %H:%M:%S")> <m<-prophet(df1)> <future<-make_future_dataframe(m,periods = 30)>

Thank you. @face @tcuongd

[amw5g]

By default, the make_future_drataframe() method assumes daily frequency. So you produced 30 records, each representing a day at 24 hour intervals beyond your last observation. I think you need to adjust the future<-make_future_dataframe(m,periods = 30) to something like future<-make_future_dataframe(m,periods = 30, freq = 60 * 15) I don't have R installed to test if I got the freq param right, but I'm referencing the example provided in the docs at https://facebook.github.io/prophet/docs/non-daily_data.html