AI/ML Workbench
A no-code machine learning environment for generating predictive insights from process data. Build, train, validate, and deploy predictive models without data science expertise.
The AI/ML Workbench enables users to build, train, validate, and deploy predictive models without requiring prior expertise in data science or machine learning. By leveraging historical process data, organizations can uncover patterns, predict future outcomes, and proactively address operational challenges.
Some common use cases include:
- Predicting the next activity in a process
- Identifying potential process violations
- Forecasting SLA breaches
- Predicting process delays and flow times
- Improving process efficiency through data-driven insights
- Supporting KPI creation and optimization
The AI/ML Workbench has been designed with simplicity in mind, allowing both business users and analysts to build predictive models through an intuitive step-by-step workflow.
Creating a Predictive Model
Click Create. Model creation has three steps:
- General: Provide Model Name and select the Event Log (only Event Logs from the Event Log Upload module are available).
- Training: Select Predictor Type (Next Activity), Input Variables (Activity Name, User, Department, Vendor, Region, etc.), arrange activities via drag-and-drop, set Testing Data Size (0.0–1.0), and select Algorithm.
- Validation: Review all configurations. Click the Edit icon to modify any section, then click Save Model.
ℹ️ The selected Event Log cannot be modified after model creation begins.
Available Algorithms
| Algorithm | Notes |
|---|---|
| Decision Tree | Transparent, easy to interpret |
| Random Forest | Generally strong default choice for most process prediction scenarios |
| XGBoost | High performance for structured data |
| K-Nearest Neighbors (KNN) | Good for pattern-based predictions |
Executing a Model
Creating a model only saves its configuration. To begin training: go to List of Models → click Edit for the desired model → click Execute.
Reports
Once execution completes, access Reports via Actions → Report. Available tabs:
- Classification Report – Accuracy, Precision, Recall, F1 Score, Support.
- Predictions – For each case: current process state, predicted next activity, prediction results.
Click Download to receive reports via email.
Model Versioning
Create multiple versions to experiment with different configurations. To create a new version: Edit an existing model → modify configuration → Save and Execute. Previous versions are preserved for comparison.
