Predictive Analytics for Direct Marketing



hello I'm Todd Jaffe director of rapid miner University I'd like to welcome you to this demonstration of rapid miner for our tour of rapid miner let's work through a real business problem and see what the end result might be from an advanced analytics process used to solve that problem in our example we will assume that we sell shaving equipment and would like to predict which people are likely to grow beard in the future today we want to select candidates for a direct marketing campaign geared towards the specialized grooming needs of beard growers we begin at the home screen for rapid miner studio you can see that we have a button to launch the application wizard you can use this wizard as an accelerator to quickly and easily realize results in response to some common business needs sometimes the easiest way to learn how to do something is to see what your destination is first to give context on where you're going while you're getting there in the case of our direct marketing campaign we would like to know where we can best spend our resources to get the most value so let's start with the direct marketing wizard on this screen we can drag and drop into this area or select a file for example an Excel spreadsheet let's go ahead and grab a data set with information about our potential customers we see that our data set consists of multiple columns the column that we are most interested in is the one we'd like to predict growing beard we can select this column with a simple click we will now create a predictive model using the other columns as input to predict the value of the growing beard column this is important because we only have information about beard growing for some of our prospects for the other rows the values in the growing beard column are missing these are the values we want to know and that our model will help us with rapid miner we'll use the examples where the values are known to build a model which predicts the value for those cases where it's unknown let's go ahead and run our analysis the results are explained to you in a guided walk through the output we start with a table containing the predictions for all prospects where the value for growing beard was missing they are sorted by our confidence that they are going to grow a beard according to the patterns derived from our data you can follow the guide and now inspected decision tree which illustrates a hierarchy of attributes and benchmarks that help us find our most likely responders next rapid miner shows the most important influence factors weighted by their impact the shaver type for example is more important than the fact that a prospect is a reader of the magazine beards today these two graphs the tree and the bar provide insight about where we want to focus our attention and our resources and perhaps equally important where we don't you can continue through the guided walk to explore the visualizations that rapid miner creates for you what this wizard ultimately shows you is a recommendation based on its analysis of our data here it suggests that aiming our campaign at a specific smaller portion of our list would give us a response rate over double that of the entire data set that fits right in with our goal of getting the best ROI from our budget this is a lot of information already which we came by pretty easily but our tour does not stop here you could choose to export our results to share this information helping to launch a targeted campaign that reaches out to that segment with a projected higher response rate let's see how simple that is we now have the option to create an HTML page let's save it and call it campaign here we see a web page that not only contains our input data but also shows us the results of our analysis in particular it gives us the records of the specific people in a recommended target group we have also saved the graphics we generated in rapid miner getting back to our tour of rapid miner besides exporting these results you might want to optimize or integrate this analysis this accelerator we have just seen can be thought of as a starting point with rapid miner you can always look under the hood and inspect the underlying analytical process this brings us to our design perspective the data is flowing from left to right in this process each building-block performs a specific analytical function this is like the square root key on your calculator performing that operation when pressed you can combine the building blocks to build your analytical solution based on those standard operations the green box on the right calculates the decision tree that we have seen everything which is delivered to the port's at the right is shown to the user or can be used by other programs if you want to integrate the processes you will build all right now that we've seen an example of what we can achieve let's see how easily you can build your own process first let's start with a clean slate and a new process next let's drag a data set onto our blank canvas and see what we have local data growing beards numbers this creates an operator that represents the loading of this data the next step is to connect the output ports of the operator to the result ports on the right side so that we can inspect the results we can now execute this process by clicking on the play button at the top alternatively we could also deploy it on a rapid miner server and execute it there we could schedule it or we could transform it into a web service for integration into our business process let's press play and see what we have for data this is a different data set containing numerical data describing our prospects we are still interested in predicting who is going to grow a beard and who is not click the statistics tab on the left to see that basic statistics are automatically available for you including average values counts and visualizations of value distributions you can click on each bar to get more information about the columns of the data set I'll click on open chart or the charts tab brings you to an extensive set of data visualizations those can help you explore data to find first patterns these visualizations are also available in the web interface of the server this parallel chart represents each case as a line and your task is to differentiate between blue lines we'll grow a beard and red lines will not impossible a different visualization might give better hints now we can see that the two classes beard and no beard actually differ in certain regions around some of our attributes for example attributes 11 45 and 36 let's use a predictive algorithm to figure out the details about those differentiators back in the process design perspective we can add one of hundreds of modeling techniques to the process and re execute it afterwards for example a decision tree each operator is explained in a help screen where the functionality expected input and all parameters are discussed this help window can be expanded or contracted as the need arises now let's see what we can learn about our data the result of the process is a decision tree decision trees are popular predictive analytics algorithms because they not only tell you what the important factors are 11 45 and 36 but also determine the optimal thresholds for decision here here and here now let's go back to the design perspective and learn more about how rapid minor will support you to create optimal processes let's say that we wanted to replace the decision tree operator with another algorithm perhaps another kind of tree algorithm like ID 3 the software will analyze in the background if this function can be applied and marks problems clearly in the process we see that this kind of algorithm is looking for a different type of input than we are providing but rapid minor can do even more than just detect problems the tool can make recommendations about how to fix them in the best way those recommendations are based on the best practices of our user community each quick fix will automatically add the necessary transformation steps you can see that rapid minor adds a new operator to implement the data transformation fix we selected the ease and use and transparency of the process flow make rapid minor and ideal tool for collaboration we've all been in situations where there is a meeting in front of a whiteboard followed by a team going off to code something after a period of time sometimes a long period results come back and either aren't quite what was intended or our spot on but generate another set of questions and the cycle starts over imagine instead that you could use that meeting to layout that whiteboard process in rapid minor and then press play how much time would that save you how much quicker could you iterate through your analysis to get key business insights how much faster could you bring those insights to market and how much sooner could you generate real value let's take a look at another feature of rapid miner the ability to save processes and run them later here we have an example of a saved process that creates a lift chart a double-click loads this process into the main process view press play and we get our chart a lift chart shows us which examples we expect to match our target and also a cumulative effect of those examples in this data set we can see that the first 10 or 15% of our population can expect to get us more than 80 or even 90 percent of our goal we didn't even have to know anything about advanced analytics in order to get this result find it load it press play done back in the design view we can reuse this process for another similar purpose simply by switching our data set that's another added value from the clarity of rapid minor since it is easy to understand how to use rapid minor more people in your company can do more things with it that means you can both empower decision-makers to perform simpler analysis tasks on their own and reduce the workload on your experts leaving them time to work on the phoning your problems so how do you get people up to speed quickly we've already seen the extensive online help built into the program but the easiest start is actually through our animated tutorials just follow the steps to quickly become an expert follow the text balloons to learn about rapid miner when you see a cartoon panel follow the guidelines within the picture to see how to perform a step in the process here we see sample data deals drag and drop so let's do that sample data deals drop here drag and drop tells you that this creates an operator tells you about the input and output ports and tells you where to find more operators next modeling classification tree induction decision tree that's what we had open before drop here it tells you to connect shows you how connect this port to this port and this port to results so let's do that output connect connect and connect now shows the play button and tells you your first process press play tells you about the results perspective and tells you to switch back to the design perspective and that's it you're done these tutorials will show you the mechanics of rapid miner and how to create predictive models with a few clicks they also cover how those models can be applied to new and unseen data and how the prediction accuracy can be determined okay by this point we have used some of the fun allottee within rapid minor studio to work through some use cases solving business problems next we will discuss some advantages of moving beyond a single analyst with an isolated desktop and what steps you can take to leverage those advantages with rapid minor server thank you

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *