Smarter Care Workshop

Transcription

1 Smarter Care Workshop 2 P a g e SPSS Lab Workbook

2

3 Overview: In this exercise, you will learn how to build a simple stream and model using SPSS Modeler. SPSS Overview: The first step in this process is to open SPSS Modeler and understand how to build a simple classification model using the automated features of SPSS. Classification models are used for many purposes so for this first exercise we will build a simple classification model to predict a machine failure. To start go to your Windows program menu and selecting IBM SPSS Modeler 16.0 as follows: Once this is done the application should open and you should see the modeler interface as shown below: 1

4 The basic metaphor of SPSS is that of a design canvas that you can drag and drop icons (nodes) on to and build a visual data mining workflow. From the top of the application, we can see a list of menus on the top and on the bottom we can see a set of tabs with icons (nodes) below each tab. We will use these nodes in this section to build a simple model. We are going to build a simple model to predict is an asset is likely to fail or not using a auto-classifier node. First click on the Sources tab and double click on the var. file node. You can also drag and drop this node on the canvas. This node allows us to pull data in from a delimited file (typically comma) to supply data to the downstream nodes. We need to associate this node to the data source. 2

5 To do this double click on the var. file node. This action will open up a dialog box and allow us to enter a file path for the source file. Using the button locate the ShredderHSRawHigherLevelGrainIII.csv file found in the following directory: C:\Healthcare_Workshop Note we can view the data from this source by clicking on the preview button (top left) to view the first ten records from the source. 3

6 Click Ok Next drag and drop a type node from the Field Ops tab on to the canvas. The result should look as follows Connect these two nodes by clicking on the first node (ShredderHSRawHigherLevelGrainIII) and hitting the F2 key and then connecting to the type node.the result should look like the following. This is known simply as a stream. 4

7 Double click on the Type node to open it up an configure the node. Ensure you are on the Types tab of the Type node and using the drop down menus for the Role configure the Type node as follows: 5

8 What this step does is to prepare the data for the downstream method/algorithm that will be consuming the data. Additionally, we are specifying a target field (FAIL) which is what we want the model to do i.e. predict failures and we are setting this field as a flag as shown below. Check to ensure Measurement for Fail is set to Flag (it could be set to continuous), click on Fail Measurement column to change if necessary. Click OK Our next step is to add a modeling node to the stream. We do this by dragging and dropping the AutoClassifier node from the Models tab on to the canvas and connecting it to the Type Node. Alternatively, you could double click the Autoclassifier node on the Modeler Pallet to add the Autoclassifier node to the last position. 6

9 At this point we have sourced, typed and connected the data to the algorithm.we are now ready to create our first model. We do this by right clicking on the AutoCluster node and choosing run from the drop down menu options. This node will pick the best algorithms to use to classify failures based on those that are most accurate. This will create a model nugget as shown below: Alternatively, click on the nugget then double click table node to add and connect it automatically. 7

10 In order to look at the output of the model we need to a bit of reordering and output the results to a table. First attach a Field Reorder node (found on the Field Ops tab) to the gold model nugget as shown. This can be done by clicking on the gold model nugget on the canvas then clicking on the Field Reorder node. This will automatically attach the Field Reorder node to the model Double click on the Field Reorder node on the design canvas and open it. Using the drop down control (far right middle) choose all of the fields from the Select Fields dialog box. This is done by holding down on the <Ctrl> button and choosing one field at a time. Then clicking on the Ok button. 8

11 When you are done you should see the following in the Field Reorder dialog box. 9

12 Using you mouse, choose the $XF-FAIL and $XFC-FAIL attributes as shown: Using the Move Up selection arrow shown below, move these two attribute up so that they are positioned below the AsseID attribute as shown below: 10

13 Click OK. Next we add a table node to the Field Reorder Node. We do this clicking on the Field Reorder node and then going to the output tab and double clicking on the table node (found on the Output tab) Your design canvas should look like the following: 11

14 We can right click on the table node and choose the Run command from the menu to produce the following output. If we scroll to the left of this output we will see two new columns $XF-FAIL which is the models prediction 1 = Fail 0 = Not Fail and a confidence value for that predictions $XFC-FAIL ( values range between 0-1). Thus our model predicts that the second asset is not likely to fail with a 99.6 % confidence. 12

15 We may also want to understand the accuracy of this model. To do this, we click on the gold model nugget and then double click on the Analysis node also found under the Output tab as shown: Right mouse click on the Analysis node on the design canvas and choose the Run menu option this should yield a dialog box as shown below. From this, we can see that the overall accuracy of this model is 97.2% 13

16 Now that we have built and verified the accuracy of our model we are going to use that model to score real data. To do this we will do the following steps: Add a second (new) var source node and connect it to the SampleData.csv file found in the following directory C:\Healthcare_Workshop as shown below: 14

17 Click on the Type node in the training stream and copy the node and paste it next to the new source node. This can be done using the <CTRL C> and <CTRL V> keys. Drag the Type node down and connect the copied Type node to the new source node as shown: 15

18 Perform the same procedure for the Gold model nugget. 16

19 Click on the second model nugget and go to the output tab and double click on the table node so that it gets added to the Scoring stream as shown: 17

20 Right click on the new table node and choose the Run menu option and execute the stream. This simulates how we would use our model to make predictions on a new data set. You should see results as follows: 18

21 We have just built our first model to predict failures. This is known as a classification model. In our next lab, we are going to develop a model to predict 30 day readmissions. Go to the File Menu and select the menu option Close Stream. You will be asked if you want to save the changes. Select the No button 19

22 In this next lab, we are going to build a model to predict those patients who are likely to return to the hospital 30 days after being discharged. The first step is to close the existing classification model we built. This is done as follows: Go to the File Menu and select the menu option New Stream. You should then have a blank design canvas as shown below. Drag and drop a variable source node from the sources tab on to the design canvas. 20

23 Double click on the variable file node on the canvas and open the Edit dialog box as shown 21

24 Using the button to the right of the file textbox go to the following directory and select the following file: C:\Healthcare_Workshop\Readmissions.csv Click the Apply button and then the Ok button. This will close the dialog box. Now connect a Type node to the variable file node as shown below. Again you can do this by clicking on the Readmissions.csv source node then double clicking on the type node found on Fields Ops tab. 22

25 Double click on the Type Node to open the Edit dialog box and click on the Read Values button. This will instantiate the measurement and value entries in the node Using the dropdown list on the Readmitted field set the measurement value to type flag as shown 23

26 Using the Dropdown list on the Readmitted field set the Role to Target as shown: Click the Apply then the OK buttons on the Type dialog box.at this point your design canvas should look like the following: 24

27 Click on the Modeling tab at the bottom of modeler and double click on the C5.0 Node. This will attach a classification node to the Type node. 25

28 Right click on the C5.0 node and using the menu drop down select the menu selection Run. This will cause the C5.0 algorithm to run and produce a model nugget as shown below. Double click on the Model Nugget. Note frame that contains the Predictor Importance. This frame details those attributes that are most predictive of a person being readmitted within 30 days. Also note the right frame which details a set of rules that allow us to determine using a few simple rules which patients are most likely to be readmitted. This can be very useful in terms of training Discharge clinicians a few simple questions to ask a person before they are discharged to determine if they are likely to be readmitted. Close the nugget dialog box using the red X in the upper right. 26

29 Click on the gold model nugget in the design canvas. The double click on the Field Reporder node found under the Field Ops tab. This should add the Field Reorder node after the Readmitted model as shown: Double click on the Field Reorder node on the design canvas and open it. Using the drop down control (far right middle) choose all of the fields from the Select Fields dialog box. This is done by holding down on the <Ctrl> button and choosing one field at a time. Then clicking on the Ok button. 27

30 When you are done you should see the following in the Field Reorder dialog box. Using you mouse, choose the $C-Readmitted and $CC-Readmitted attributes as shown: Using the Move Up selection arrow shown below, move these two attribute up so that they are positioned below the Readmitted attribute as shown below: 28

31 Click OK. Next we add a table node to the Field Reorder Node. We do this clicking on the Field Reorder node and then going to the output tab and double clicking on the table node (found on the Output tab). Your design canvas should look like the following: 29

32 Right click on the table and select the menu item Run. Using the horizontal scroll bar slider, scroll to the far left so that you can see the $C-Readmitted and $CC-Readmitted columns. These represent the prediction of if a person is going to be readmitted along with the confidence in that prediction. You can now close the Table window. You have successfully built a simple model to predict 30 day hospital readmissions. Close the table view by clicking on the red x in the top right of the table. We may also want to understand the accuracy of this model. To do this, we click on the gold model nugget and then double click on the Analysis node also found under the Output tab as shown: 30

33 Right mouse click on the Analysis node on the design canvas and choose the Run menu option this should yield a dialog box as shown below. From this, we can see that the overall accuracy of this model is 95.97%. 31

34 Close the Analysis dialog box. We can save the models we create using the following process. Please note : If you have a memory stick or flash drive you may want to use an alternate directory path that the one detailed below. Go to the File Menu, choose the Save As menu option. This will bring up a dialog box as shown: In the File Name textbox change the default name to Readmissions_Solution as shown and ensure that it is being saved in the following directory path:c:\healthcare_workshop. Click on the Save button. Go to the File Menu and select the menu option Close Stream. In the next lab, the instructor will guide you through the use case (predicting patient complications) and the basic elements of the stream. The first stream we will examine is the Complications_Prediction2.str stream. This stream can be found in the following directory: C:\Healthcare_Workshop\Costs-LOS-Models\HC_Streams\HC_Streams 32

35 To open this stream go to the File menu in Modeler and choose Open stream this will allow you to go to the correct directory and pick the Complications_Prediction2.str file. Once open the stream should look like the following: The instructor will now guide you through the stream and its general functionality/purpose. After the discussion is complete, close the stream. This is done by going to the File Menu and select the menu option Close Stream.In the next lab, the instructor will guide you through the use case (predicting expected patient costs) and the basic elements of the stream. The stream we will examine is the Costs_Predictions_Including_Procrdure.str stream. This stream can be found in the following directory: C:\Healthcare_Workshop\Costs-LOS-Models\HC_Streams\HC_Streams To open this stream, go to the File menu in Modeler and choose Open stream this will allow you to go to the correct directory and pick the Costs_Predictions_Including_Procrdure.str file. 33

36 Once open the stream should look like the following: The instructor will now guide you through the stream and its general functionality/purpose. After the discussion is complete, close the stream. This is done by going to the File Menu and select the menu option Close Stream. Hopefully this short SPSS Healthcare workshop has helped you better understand how you can use SPSS Modeler to predict many future healthcare type events. These include areas such as readmissions, costs, length of stay, high risk pregnancies, fraud, complications. 34

37 NOTES: 35

38 36

39

40 Copyright IBM Corporation The information contained in these materials is provided for informational purposes only, and is provided AS IS without warranty of any kind, express or implied. IBM shall not be responsible for any damages arising out of the use of, or otherwise related to, these materials. Nothing contained in these materials is intended to, nor shall have the effect of, creating any warranties or representations from IBM or its suppliers or licensors, or altering the terms and conditions of the applicable license agreement governing the use of IBM software. References in these materials to IBM products, programs, or services do not imply that they will be available in all countries in which IBM operates. This information is based on current IBM product plans and strategy, which are subject to change by IBM without notice. Product release dates and/or capabilities referenced in these materials may change at any time at IBM s sole discretion based on market opportunities or other factors, and are not intended to be a commitment to future product or feature availability in any way. IBM, the IBM logo and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol ( or ), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at Copyright and trademark information at ibm.com/legal/copytrade.shtml Other company, product and service names may be trademarks or service marks of others. 38