Unanswered

Have I built my dataset correctly or am I completely wrong?

(0) Share

Report

Posted on by barryjarvis

Hi all,

I've never done any machine learning or ai stuff before, but I'd really like to dive into AI Builder and see if I can use to it to help my company.

I would like to answer the question "Will this customer renew their maintenance contract?" with a AI Builder prediction model.

After reading some of the MS documentation, I've come up with the below affecting factors:

cost of current contract
current contract type
length of existing contract
how many times have they renewed previously
how many engineers visits
number of technical cases
how many assets on contract
do they have any other contracts
is the contract via a reseller
historic outcome = contract status is cancelled

All of this data is available to me via our Dynamics CRM, and would translate to a table like so:

contract cost	contract type	contract length (days)	previous renewals	visit count	case count	asset count	other contracts	via reseller	outcome
1200	premium	365	2	1	0	2	N	Y	Cancelled
900	standard	365	0	2	3	1	N	N	Renewed

My questions are:

Am I heading the right direction - I've never done anything like this before, so I'm in the dark a bit.
If I had a few hundred records of the above data, would that train a model to predict the outcome of our future maintenance contracts?
I believe I need to include the historic outcome in the training dataset so that the model can learn what factors influenced that outcome? - this sounds really obvious but I need to check that I'm thinking about it the right way

Any help on this would be hugely appreciated.

Categories:

AI Builder

I have the same question (0)

All responses (2)

Answers (0)

Sort by

CedrickB Microsoft Employee on at

Like (1)

Report
Copy link

Link copied!

Hi,

1. Yes, you are in the right direction. 🙂

2. 100 records is really small. 1 000 would be better and generally the more data you provide to the model the better the prediction will be.

3. Here is a simplified walkthrough:

a. Make sure to copy your table data to an entity in Common Data Services

b. Add also the new data to be predicted in this entity. The field "outcome" should be either empty or with a specific OptionSet value (E. g. "To be predicted")

c. Create an AI Builder "Prediction" model

d. In the wizard, choose "outcome" as your learning outcome field and selected in the Matrix the OptionSet value for the rows to be predicted

e. Go through the wizard, train and publish your model

Three new columns will be added to this entity: predicted the value predicted, confidence how confident the model is about this prediction and explanation why the model predicted this value.

A batch process will soon be triggered and will fill in values for these rows to be predicted.

Please refer to the documentation for more details https://docs.microsoft.com/en-us/ai-builder/prediction-create-model

Was this reply helpful? Yes No
barryjarvis on at

Like (0)

Report
Copy link

Link copied!

@CedrickB thanks for the reply 🙂

One thing I'm not getting my head around fully is the 'outcome'.

My understanding was that I would include the outcome of the historic records that we're using to train the model on. i.e. the contract was cancelled, or it was renewed.

My data set only has the outcome options of 'cancelled' or 'renewed' (so there's no worry about using an OptionSet), there is no option of them being blank?

Am I not right in saying for training the model the historic outcome would need to be either one of the above options? So that the model can learn what factors affect each outcome?

If the outcome is blank then how does the model know what factors affect it?

Obviously going forward (after training the model) I would expect the outcome to be blank and the model to predict it, but I'm specifically talking about training the model at the moment.

Sorry if this is a dumb question - just trying to get my head around it?

Was this reply helpful? Yes No