This paper introduces Nirdizati: A web-based application for generating predictions about running cases of a business process. Nirdizati is a configurable full-stack web application that supports users in selecting and tuning prediction methods from a list of implemented algorithms and enables the continuous prediction of various performance indicators at runtime. The tool can be used to predict the outcome, the next events, the remaining time, or the overall workload per day of each case of a process. For example, in a lead-to-order process, Nirdizati can predict which customer leads will convert to purchase orders and when. In a claim handling process, it can predict if a claim decision will be made on time or late. The predictions, as well as real-time summary statistics about the process executions, are presented in a dashboard that o↵ers multiple visualization options. Based on these predictions, process participants can act proactively to resolve or mitigate potential process performance violations. The target audience of this demonstration includes process mining researchers as well as practitioners interested in exploring the potential of predictive process monitoring.
Predictive Process Monitoring [
As an example, consider a business process for managing loan applications; the applicant can be advised about the combinations of loan amount and duration that are the most likely to lead to an acceptance of the application, given the history of the application until now and the personal data of the applicant (e.g., age, salary, etc.).
Several approaches have been proposed in the literature to tackle common predictive process monitoring tasks. However, so far, these approaches have largely remained in the academic domain and have not been widely applied in real-time scenarios where users require a continuous predictive support.
In this paper, we present Nirdizati, a pioneering open-source web-based predictive monitoring tool, which is able to fill this gap between research and practice, by providing business analysts with a highly flexible instrument for the selection, generation and analysis of di↵erent predictive models, and end-users with continuous runtime predictions.
Nirdizati consists of two components: Nirdizati Training and Nirdizati Runtime (Fig. 1). Nirdizati Training takes as input a business process event log and produces one or more predictive models, which can then be deployed in Nirdizati Runtime. Once a model is deployed, Nirdizati Runtime listens to a stream of events coming from an information system supporting the process, and produces a stream of predictions. These predictions are then visualized in a continuously updated web dashboard. 2
Nirdizati Training is the component of Nirdizati that allows users to produce predictive models later used by Nirdizati Runtime for making predictions on a stream of events. It provides several algorithms for generating predictive models suitable for di↵erent types of predictions and tailored to each specific dataset. For example, it is able to build predictive models for predicting remaining time, the next activity to be performed, whether a certain outcome will be achieved or not and the overall workload per day. To this aim, the training component of Nirdizati relies on two phases: a training and a validation phase. In the former, one or more predictive models are built; in the latter, their suitability to the specific dataset is evaluated, so as to support the user in selecting the predictive model that ensures the best results.
Nirdizati Training is composed of a front-end application (Fig. 2), which allows users to select the prediction methods and to assess the goodness-offit of the built models, and a back-end application responsible for the actual training and validation. The back-end application is, in turn, composed of four submodules shown in Fig. 3. A Log Manager is in charge of managing the logs. Uploading and retrieving the logs are the basic operations of this module. The Encoder is responsible for parsing the logs, labeling them according to the desired type of prediction, and preparing the data for the training phase. In the Training
Information system process events prefix col ator event prefixes TKeaxftka processor Kafka topic TeSxttream oTfeJSxOtN objects case outcome predictor slow probability case outcome predictor reject probability remaining time predictor predictions join events to predictions events with predictions
Predictive dashboard submodule, the encoded data is split into training and validation set used for evaluation purposes, and the predictive models are built from the training data. Finally, the Evaluation submodule tests the validation data against the created model(s) to get accuracy measures with respect to the ground truth, available from the complete traces in the validation data. The back-end application comes with a storage module for saving the uploaded logs and the generated predictive models. 3
Once the predictive models have been created, they are used by the Runtime component to make predictions on ongoing cases. Nirdizati Runtime takes a stream of events produced by an information system, transforms it into a stream of predictions, and visualizes those predictions in a web-based dashboard. The transformation is implemented using the dataflow pipeline in Fig. 4. The pipeline is built on top of the open-source Apache Kafka stream processing platform. The “predictor” components of the pipeline are the predictive models from Nirdizati Training. The “topic” components are network-accessible queues of JSON messages with publisher/subscriber support. This allows the computationally intense work of the predictors to be distributed across a cluster of networked computers, providing scalability and fault-tolerance. The “collator” component accumulates the sequence of events-to-date for each case, such that the prediction is a stateless function of the trained predictive model and of the case history. This statelessness is what allows the predictors to be freely duplicated and distributed. The “joiner” component composes the original events with the various predictions, ready for display on the dashboard.
The dashboard provides a list of both currently ongoing cases (colored in gray) as well as completed cases (colored in green), as shown in Fig. 5. For each case, it is also possible to visualize a range of summary statistics including the number of events in the case, its starting time and the time when the latest event in the case has occurred. For the ongoing cases, Nirdizati Runtime provides the predicted values of the performance indicators the user wants to predict. For completed cases, instead, it shows the actual values of the indicators. In addition to the table view, the dashboard o↵ers other visualization options, such as pie charts for case outcomes and bar charts for case durations.
Process workers and operational managers – typical users of Nirdizati – can set some process performance targets and subscribe to a stream of warnings and alerts generated whenever these targets are predicted to be violated. Thus, they will be capable of making informed, data-driven decisions to get a better control of the process executions. This is especially beneficial for processes where process participants have more leeway to make corrective actions (for example, in a lead management process).
Nirdizati is a configurable full-stack web application that supports users in selecting and tuning prediction methods from a list of implemented algorithms and that enables the continuous prediction of various performance indicators at runtime. The predictions are presented in a dashboard that o↵ers multiple visualization options.
A video demo of Nirdizati can be found at http://youtu.be/0nr14lX04-I. A public release of Nirdizati Training and Nirdizati Runtime is available at http://training.nirdizati.com and http://dashboard.nirdizati.com, respectively. The source code is available under the Lesser GNU Public License (LGPL) at http://github.com/nirdizati.