Year: 2026
Type: conference-paper
Abstract: Integrating data analytics and time-series forecasting into IoT applications commonly forces engineers to maintain preprocessing, training, evaluation, visualization, and runtime glue code outside the system model. ML2++ addresses this separation by extending ML-Quadrat with declarations for datasets, feature roles, preprocessing, lag windows, forecast horizons, algorithms, metrics, plots, and bindings from predicted values to IoT behavior. This paper focuses on the tooling extensions and implementation evidence beyond the previously published language overview. It presents an Xtext-based textual editor, a Sirius Web graphical studio, a browser-based Quick-Start execution environment, normalization of exported graphical models into the shared EMF/Ecore representation, a common downstream code-generation path, and a Dockerized Orchestrator for generated Python/Java projects. The demonstration follows a river-flow model from sensor ports and messages through multi-step forecasting to an illustrative alert decision. Three use cases exercise seven positive-path configurations spanning recurrent, statistical, and tree-based forecasting. The accompanying artifact provides model specifications, aggregate reported metrics, reference workflows, tests, and a machine-readable account of the evidence currently available and missing.
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7220443133