Alteryx Workflow - Predicitve Model

[Karthik_7694]

Hi All,

I have a dataset for which I need to build a Linear Regression Model using Predictive Module in Alteryx.

I hereby attach the dataset for your reference. 

Use case :

Predicting Catalog Demand

The task is to predict how much money the company can expect to earn from sending out a catalog to new customers. Need to build a linear regression model and apply the result to the mail-order catalog business problem

The two data files are: p1-customers & p1-mailinglist.

We need to predict the expected profit from the 250 new customers in the mailing list file. Unless we can expect a profit of >$10,000, we will go ahead with printing the catalogs to send to these new customers. So because we are data rich, and trying to solve a numerical continuous problem, we will be using the linear regression model.

It will of great help if some one guides me how to buid the workflow in alteryx and also can some one guide me once we develop this model in alteryx how to test the predictive model with new set of dataset and also how to push this workflow in alteryx server/gallery.

Thanks,

Karthik

[apathetichell]

Hi... Are you missing your dependent variable or is score_Yes your dependent variable? If so it looks like it's kind of designed for a 1/0 true false. Also in your test data since you are building your model on it - you kind of need to decide if it's a 1 (yes) or a 0 (no)?

this is probably better suited for logistic regression or another yes/no style regression model (tree model?)... as a linear regression it's not giving me much...

Report for Linear Model Linear_Regression_4

2

Basic Summary

3

Call:

lm(formula = Score_Yes ~ Customer_Segment + City + Store_Number + Avg_Num_Products_Purchased + X._Years_as_Customer, data = the.data)

4

Residuals:

5

Min	1Q	Median	3Q	Max
-0.2375	-0.1152	-0.0659	0.0579	0.6975

6

 Coefficients:

7

Estimate Std. Error t value Pr(>|t|)   (Intercept) 0.3977648 0.59080 0.67326 0.50147   Customer_SegmentLoyalty Club Only 0.0076910 0.03021 0.25460 0.79927   Customer_SegmentLoyalty Club and Credit Card -0.0083786 0.04679 -0.17908 0.85804   Customer_SegmentStore Mailing List 0.0280017 0.05308 0.52752 0.59836   CityAurora -0.0339467 0.04762 -0.71283 0.47669   CityBroomfield 0.0392557 0.06591 0.59562 0.55203   CityCastle Pines -0.1718996 0.19680 -0.87349 0.38333   CityCentennial -0.0284672 0.07637 -0.37276 0.70968   CityCommerce City -0.1130052 0.14069 -0.80321 0.4227   CityDenver -0.0320047 0.04700 -0.68090 0.49664   CityEdgewater 0.6482706 0.19597 3.30793 0.00109 ** CityEnglewood 0.0307923 0.10522 0.29266 0.77005   CityGolden -0.0962932 0.14104 -0.68274 0.49548   CityGreenwood Village -0.0856867 0.19848 -0.43172 0.66636   CityHighlands Ranch -0.0605954 0.14340 -0.42255 0.67303   CityLakewood 0.0228237 0.05367 0.42524 0.67107   CityLittleton 0.0136295 0.07804 0.17464 0.86152   CityLouisville 0.0165866 0.19656 0.08438 0.93283   CityNorthglenn 0.1045096 0.11809 0.88501 0.3771   CityParker 0.0506121 0.10634 0.47593 0.63458   CityThornton 0.0485668 0.11772 0.41256 0.68033   CityWestminster 0.0080556 0.08895 0.09056 0.92792   CityWheat Ridge 0.0225780 0.09442 0.23913 0.81122   Store_Number -0.0002043 0.00551 -0.03707 0.97046   Avg_Num_Products_Purchased 0.0009407 0.00561 0.16768 0.86698   X._Years_as_Customer -0.0638804 0.04095 -1.55994 0.12019

Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

8

Residual standard error: 0.19097 on 224 degrees of freedom

Multiple R-squared: 0.09166, Adjusted R-Squared: -0.009722

F-statistic: 0.9041 on 25 and 224 degrees of freedom (DF), p-value 0.5998

9

Type II ANOVA Analysis

10

Response: Score_Yes

Sum Sq DF F value Pr(>F)   Customer_Segment 0.01 3 0.11 0.95421   City 0.69 19 1 0.46678   Store_Number 0 1 0 0.97046   Avg_Num_Products_Purchased 0 1 0.03 0.86698   X._Years_as_Customer 0.09 1 2.43 0.12019   Residuals 8.17 224

Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

[KilianL]

Hi @Karthik_7694 ,

Great to see a predictive use case!

I created an example predictive workflow going over data health, picking variables and testing the model on a validation dataset.

I used the customer list data to calculate spend, prepared the data for the predictive model, split the data into training and validation data set and then used a linear regression to predict spend. The score tool in the end uses the model to predict spend for the validation dataset. Here you can compare actual spend with predicted spend.

I didn't get good results (R-squared of ~0.5, meaning only half of the variation is explained by the model), and I think there is more data needed. Looking at the Model results, the only significant factor to predict spend was the customer segment. (using average spend is not best practice, as this includes information about the target spend)

If you have time series data, you could try time series forecasting to predict spend.

Please mark this as the solution if it answers your question, it will help others to find solutions quicker.

Kind Regards,
Kilian
Solutions Engineer - Alteryx