Binary Classification Inspector Node in Knime

in this short video we will take a look at the new binary classification inspector node in 9 this is a fairly new node that was released in the last version of 9 the past you may have used the score node along with other nodes to gauge the prediction performance of the model but with the new moon node you can do that easily all in a visual format let's take a look at the output of this

node you can see it launches a Chrome browser and the output contains a number of metrics and graphs you can take a look at the accuracy rate the error rate sensitivity visibility precision false negative rate and I'm not sure where there's no recall matrix it would be nice to have a recall matrix as well but nevertheless it's a great really good visualization tool for your model prediction performance so now the other really good thing about this node is that you can drill down into any of these prediction any of your models and you can change the threshold and this is

your prediction probability threshold so you can adjust that and see how your how the confusion matrix changes along as you change the threshold maybe you're looking for a higher sensitivity and lower specificity or higher precision versus a higher specificity you can change the threshold and get a good idea of how all of these measures will change as you change the probability cut-offs so now how do we get to this data and what's involved in using this node so that's actually very straightforward so given I'm assuming that you already have built out a model and you have you are evaluating and the performance of weights models that you have implemented and so you're using the prediction node right here I'm using three of them one for the random forest assumed tree and the gradient boosted trees now pushing and pushing the output of this node to a score I know it which we use to do that previously in the nine what we do is we work first of all we need to make sure that we are sending the class proper to ladies for each of these and so this is this will this is the metrics that the inspector node uses to measure the performance to visualize the performance so all of make sure that you are adding the class property is to all the predictor nodes this is not the default setting so you actually have to go in and do that for each of the node predictor node and I'm just making sure that I do have that set and all of them yes I do so now once you have that you can change the column name if you want to make it more reader friendly so pair I change the true value which is what we are looking for to the probability so I call it the decision tree probability and then the prediction is a binary responsible so I just use the real variable names and this just makes it easier for a reader to go through the output of the binary classification inspector node unless you have that you just join all three of or however many models that you have joined them based on the key that you're using in this case in this data set the phone number was the key so we'll keep that as it is and similarly join the two other output and then once we have that we can execute the binary classification inspector node and we should get the output but before we do that let me just take a look at let's the configuration for the binary

classification inspector the only thing that you would have to change a so obviously the target column this is your response variable that you're measuring against and also you want to add all the probabilities that came in from the previous nodes so that's why I changed the names it's easier to read and identify them right here and so this is the probability that's coming from the predictor nodes okay so these are the only two things you need to change you there are the up settings that you want to experiment with go for it and so the and the positive class in our case we want to see if it's at or if it's a turn on and not so we set it to true and the target column is true let's go ahead and click OK and that's fine and let's take a look at that so this is what we saw earlier it launches a Chrome browser and here we have a nice visualization of all the metrics so this is a really cool node that you want to explore and it makes building models and evaluating them really fun that's a key thing it has to be fun so hopefully you'll start using this node and I have already done that and finding it very useful thank you