PMML or Predictive Model Markup Language is an open-source XML-based standard developed by the Data Mining Group (DMG) to represent predictive models. If I explain it further, you can write your algorithms in any language such as Python, R, or any other application, and then build the predictive model. After that, you can create its PMML version.
Why PMML is popular among many commercial and open-source platforms?
- Interoperability: It allows you to share and reuse your models across multiple applications. You do not need to fret about having to use a specific language or tool to create a model and then connect it with your applications.
- Available offline: Since it is a single XML file, there is no need for any web interface to connect the models. Instead, you can use the file offline.
Especially, with the rise of Artificial Intelligence, predictive analytics has started playing a major role in modeling and automating decisions. Although it seems using AI/ML helps companies to make better decisions based on real-time data, but the challenge is to understand, model and define the decisions which provide business value to organizations, and then operationalize the AI/ML model as part of the decision automation scenarios within organizations.
In this scenario, PMML plays a big role to enable organizations to consume an AI/ML model as part of the automation of operational decisions.
Once the PMML model is ready it can simply be plugged into the decision models a part of the end-to-end decision automation.
To provide a better understanding, let’s discuss an example. The following snippet was taken from a UCI ML Breast Cancer Wisconsin (Diagnostic) dataset. It contains a set of data from a breast cancer screening including a column that says whether the cancer cells were detected in each test.
Based on the above dataset, we used Gaussian Naive Bayes Classifier to train and create our predictive model in Python. Therefore, our model can predict the possibility of cancer once we input patient cancer screening data. Finally, using that predictive model, we created the PMML file similar to the below snippet.
This PMML file can be used across multiple applications.
Let’s see an example of how we can use this file in a Decision Requirement Diagram (DRD).
Application of PMML as a Part of a Decision Model
To showcase the application of PMML, we have created the below model in FlexRuleTM Advance Decision Management Suite. The example takes a set of inputs from a patient’s breast cancer test and determines whether there is a possibility of cancer.
As an output, it shows whether cancer cells were detected along with the probabilities.
To see the complete tutorial, visit our Resource Hub.
Using this is output, you can take further decisions such as whether to contact the patient or doctor immediately, the type of treatments, and then create the complete decision model. Without stopping there, the output of the complete system can be a data source to a whole new data collection creating numerous possibilities to build a better system.
In a constantly changing world where the decision-making process changes, there is no doubt that we need a solution that makes decisions based on predictive models built based on recently updated data. Particularly, areas such as healthcare, finance, and education have endless opportunities to improve operational decisions outcome using the latest information and data.
That’s why PMML execution capability plays a critical role to enable decision automation scenarios to be able to consume predictive models as part of the end-to-end decision automation and helps operationalizing AI/ML projects to provide the real business value to organizations.
Published August 20th, 2020 at 03:05 pm