PPMx is a no-code platform for Predictive Process Monitoring (PPM). It provides an end-to-end pipeline from event log preprocessing to model training, explainability, and evaluation supporting Transformer and Graph Neural Network (GNN) architectures through a browser-based dashboard with no scripting required. The tool supports the publication 'PPMx: a Predictive Process Monitoring Workbench'.

🌐 Live: https://ppmx.app/

---

PPMx Architecture

Key Features

• No-code workflow for training and explainability.
• Transformer and GNN support for next activity, event time, and remaining time.
• Explainability and evaluation in one pipeline.
• Frontend dashboard powered by a FastAPI backend

Tech Stack

• Frontend: Vite + React + TypeScript
• Backend: FastAPI + Python
• ML: TensorFlow, PyTorch, PyTorch Geometric

PPMx System Structure and APIs This diagram summarizes the frontend-backend structure and the APIs used across the system.

Requirements

• Python 3.10+ recommended
• Node.js 18+ recommended

Quickstart (Frontend + Backend) Clone the repository first and then follow the steps:

1. Create and activate a Python virtual environment inside PPMx dir.

#Create Virtual environment
python -m venv .venv

# Activate Environment 
# Windows PowerShell
.venv\Scripts\Activate.ps1
# macOS/Linux
source .venv/bin/activate

2. Install dependencies.

pip install -r requirements.txt

3. Install frontend dependencies.

cd frontend
npm install

4. Configure the frontend API base URL.

# Windows PowerShell
Copy-Item .env.example .env.local -ErrorAction SilentlyContinue
# macOS/Linux
cp .env.example .env.local 2>/dev/null || true

Add the following to frontend/.env.local:

VITE_API_BASE_URL=http://localhost:8000

5. In the New Teminal (T2), Start the backend API.

uvicorn backend.main:app --reload --port 8000

6. In the previous Terminal (T1), Start the frontend dev server.

npm run dev

[Note: The program runs frontend and backend in 2 seperate Terminals (T1 & T2) Simultaneosly]

Frontend will be available at http://localhost:5173.

Starting Dashboard

Sample Datasets These sample event logs are available from the Dataset Setup step in the UI and can be downloaded from here:

• BPI_2020_Log_DomesticDeclarations.xes
  https://drive.google.com/file/d/1AzsG6xg9ftqQRSW2yyY7gaLcBK1qkSXv/view?usp=sharing
• BPI_2020_Log_InternationalDeclarations.xes
  https://drive.google.com/file/d/1Oz-Mbmlt4n6aiJlHSNUP3oVk-zH5UK82/view?usp=sharing
• BPI_2020_Log_PrepaidTravelCost.xes
  https://drive.google.com/file/d/1dSsgdl3RIetTTNcOOfzgzF3pxG5KLyAg/view?usp=sharing
• BPI_2020_Log_RequestForPayment.xes
  https://drive.google.com/file/d/1V6hbOWKbN2MtnPtOjPAgPKhqDwvYJTym/view?usp=sharing
• Sepsis Cases - Event Log.xes
  https://drive.google.com/file/d/1fzY9b5ly-SHUmYvOOMXbO510hdYRcQqq/view?usp=sharing

Usage Workflow

1. Preprocess the event log: upload raw CSV/XES or a preprocessed CSV. Optionally skip preprocessing by uploading pre-split datasets via the GUI. Map required columns (case ID, activity, timestamp; resource optional). For standardized BPI 2017/2020 logs, automatic column detection is supported in batch/CLI.
2. Train and predict: choose Transformer or GNN and select a task. Classification supports next-activity and custom target prediction; regression supports event-time and remaining-time. Configure hyperparameters in the GUI or use defaults, then run training and generate test-set predictions.
3. Explainability analysis: for Transformers, use SHAP (feature importance bar + beeswarm) and LIME (local explanation plots). For GNNs, the current path uses a PROPHET-style heterogeneous GAT with GNNExplainer-style local node/edge explanations and global view-importance aggregation.
4. Evaluation: compute faithfulness, comprehensiveness, sufficiency, monotonicity, and method agreement to compare explanation quality.
5. Export artifacts: each run is packaged as a ZIP containing the trained model, prediction CSVs, explainability plots, and evaluation summary for reproducibility.

Project Structure

• frontend/ Frontend app (Vite + React).
• backend/ FastAPI service for training and inference.
• BPI_dataset/ Sample datasets used for experiments.
• explainability/ Explainability methods and reports.
• gnns/ GNN training and prediction pipeline.
• transformers/ Transformer training and prediction pipeline.

Troubleshooting

• If you see port conflicts, change the backend port and update VITE_API_BASE_URL.
• If training is slow, reduce dataset size or use a GPU-enabled environment.

License This project is licensed under the MIT License - free to use, modify, and distribute with attribution.