August 3, 2017

Train predictive models

Authors: Stanislav Mõškovski, Simon Raboczi, Ilya Verenich, Marlon Dumas, Marcello La Rosa

Nirdizati Training plugin allows users to generate predictive models for different types of predictions. Specifically, it is able to predict generic process properties, such as remaining time until case completion, the next most likely activity to be executed and whether a case will take longer than a user-defined time threshold. Additionally, the tool can build models to predict log-specific case properties, for example, the total application cost in an insurance claims handling process.

  • As a minimum input, a user only needs to select the event log and the variable to be predicted.

  • Experienced users may switch to the advanced mode to fine-tune training configuration and even train multiple models at once.

  • Once the necessary models have been built, the tool assesses their accuracy with respect to multiple evaluation metrics using a held-out validation set.

  • Trained models can be downloaded as Python pickle objects and pushed to the Runtime component to make predictions for ongoing cases.

A screencast of this plugin can be found here.

For non-Apromore users, a stand-alone version of the plugin can be accessed at https://training.nirdizati.org