Predict or Forecast Future dates resource hours based on the Previous Values

[sheeba]

Hello,

I need to predict or forecast the next 2 weeks' hours based on the previous 2 weeks' hours as shown below. How can i achieve this?

Week of	User	Activity Type	Activity	Mon	Tue	Wed	Thu	Fri	Sat	Sun
4/6/2020	a	ABC	Training	--	2	1	--	1	--	--
4/6/2020	b	ABC	Training	--	--	--	--	--	--	--
4/6/2020	c	ABC	Training	--	1	--	--	1	--	--
4/6/2020	d	def	Example 1	--	--	4	4	4	--	--
4/6/2020	a	def	Example2	10	10	4	--	--	--	--
4/6/2020	b	def	Example 3	1	1	1	1	--	--	--
4/6/2020	c	def	Example 4	5	6	4.5	6	5.5	--	2
4/13/2020	a	ABC	Training	--	4	1	--	1	--	--
4/13/2020	b	abc	Training	--	--	--	--	--	--	--
4/13/2020	c	abc	Training	--	1	--	--	1	--	--
4/13/2020	d	def	Example 1	--	--	4	4	3	--	--
4/13/2020	a	def	Example2	7	10	6	--	--	--	--
4/13/2020	b	def	Example 3	1	1	1	1	--	--	--
4/13/2020	c	def	Example 4	5	5	5.5	6	5.5	--	2
4/20/2020	a	abc	Training
4/20/2020	b	abc	Training
4/20/2020	c	abc	Training
4/20/2020	d	def	Example 1
4/20/2020	a	def	Example2
4/20/2020	b	def	Example 3
4/20/2020	c	def	Example 4

[RolandSchubert]

Hi @sheeba ,

"predict" or "forecast" usually points to Time Series (ETS, ARIMA) or Predictive tools (e.g. Linear Regression). Unfortunatey, the available data set is very small (only two weeks of history) - tools based on previous values will not return suitable results. I don't think so either, that a regression model would fit.

My solution would be to use an average of last two weeks values as a forecast for the next week, perhaps a weighted average (e.g. last week 60%, last but one week 40%) could make sense.

I've attached a sample workflow to illustrate. Let me know, what you think about this approach.

Best,

Roland

[sheeba]

Hello Roland,

Thank you so much! That works fine. But, I wanted the 2 future dates also to be calculated.4/20/20 and 4/27/2020 also need to be calculated instead of providing as an input.

[RolandSchubert · Accepted solution]

Hi @sheeba ,

I added a Generate Rows tool to create the additional weeks. Hope that works for you.

Best,

Roland 

[sheeba]

Thank you so much, Roland!!