Decision Tree & Neural Networks - SAS Enterprise Miner

hello everyone my name is aya miss Eva in this video I would like to introduce how to use SAS Enterprise miner to build decision tree model and neural networks model and then we can compare the results and see which model is better model ok so I already went ahead and created a project called demo I also added a file import node to load the data set before we get this I would like to take a look at the data set so this is the data say that I loaded as you can see this is from a phone company and this captured data of each customer whether they have international plan or a voicemail plan or how many voicemail message say they had or how many times they call customer service so the target for us is to analyze and predict what is driving customer to walk away which is churn here indicated in binary here I already uploaded the data so I'm gonna click here and edit valuables I already changed this but as you can see I changed the term role to be target because this is what I want to know as a target and as you remember it was binary so I also changed the level to be binary and other valuables was were numeric so I left it as interval and some other stuff for a binary side when I hadn't changed also binary as well ok and the next thing I want to do is to explore the data set by using stuff Explorer this helps us learn more about the data so I'm gonna go ahead and run this so this is really nothing to me know to kind of tell us of what it is that the data say it's telling us it's a summary and then we can just get a good understanding of the data set so as you can see we can see if there is missing input which there isn't in this case and there is also me in standard etc for the whole dataset next I want to do is to partition the data into training validation and testing data so right here I already put 60 for the training data and 20 for the validation in 20 for the test and we can go ahead and partition the data here you see that we had two thousand ninety four data points in total and twelve hundred is is in the training day data set and four hundred and twenty each for the validation and in testing data set so now I want to use utility and control points drag and drop it here so what this does is to really allow me to use the same data set until this point which is fall imports that exploring data partition and then connect to the multiple models so I'm gonna go ahead and connect these in the first model I wanna do is the decision tree so I'm gonna click model and choose this isn't tree and I'm gonna go ahead and connect this so here you can see there's a maximum amount of branch which is two and theft six and category or size five you can always change this but for this video I'm gonna use the defaults so I'm gonna go ahead and run this results so here's this tree so as you can see this is the root node and it contains the whole data points and the first thing that that was in first variable that was an importance in this decision tree was state charge and as you can see if it's less than 44.78 eighty nine point forty two percent of the time customers stay with the company and 10.58 person at the time customer left the company the other important variable is customer service call so as you can see if they call less than 3.5 times to customer service calls that means 19 3.04 percent of the time customer stayed with the company whereas 6-point 96 percent of the time customer left a company such as the destiny tree is really easy for us to interpret and then that's exactly the reason why people do like to use this decision tree next model I want to do is neural networks I'm gonna go ahead and drag this down and as I mentioned I could use the same data from this control point something I'm gonna go ahead and connect those here I'm gonna click network make sure multi-layer perceptron is selected and the optimization here I'm going to change from default to backpropagation algorithm and iteration max minner ratio I wanted to 25 and maximum time I want to do 30 minutes because I don't want it to take 4 hours to process this so click hit ok and run it okay so this is a neural network diagram and there is a every Square and let me see the output there are number of iterations and I think this hi1 this is hidden node 1 and this hidden node 2 and then these are the weights that were that were calculated by this algorithm and there are whole bunch of calculation going on it's really hard for us to interpret by just looking at this result but I would like to focus on this classification table confusion matrix here we predicted the customer to not leave 0 negative and then we got it right for 1056 and it should positive we predicted the customer to leave and then 135 points it was actually correct correctly predicted in those 22 and 43 were miss classification now I would like to go ahead and compare these models and see which model actually did better so here I'm gonna connect these two model comparison and click run ok so this is the result so I'm gonna go ahead and look at this fit statistics here you can see this is not miss classification rate and as you can see this isn't read a way better than the neural network so we can conclude that this is entry what's the better model so this is it for the demo I wanted to do today thank you so much for watching

Channel: Ayame Shiba

Views: 3,107

Rating: 5 out of 5

Keywords:

Id: om02_lChPHc

Length: 7min 51sec (471 seconds)

Published: Sat Apr 08 2017