Blueprint: Machine Learning & Predictive Analytics

### Description

For a given data stream e.g. numerical data published via MQTT Topic, what can we offer in the sense of predictive analytics.

### Packaging

- Could be available within the FlowFuse UI, or available as a pre-packaged FlowFuse node in the Node-RED Editor experience, such that developers/users can use it however they see fit.

#### Forecast Node

The idea would be such that we could easily project values moving beyond known data in a real-time scenario.

We also need to be clear on any clear error bars/boundaries

### "Traditional" Options

- Exponential Smoothing: https://otexts.com/fpp2/expsmooth.html
- [Holt-Winters](https://otexts.com/fpp2/holt-winters.html) -

### Machine Learning Options

- TensorFlow.js: Would require separate "Train" and "Predict" workflows/nodes.
-

###October 2025 edit

With the release of our MCP and ONNX nodes, we are now very well-positioned to offer a predictive analytics solution for FlowFuse Enterprise customers. Our customers are (1) already collecting significant amounts of data from machines and sensors, (2) we have Tables and Dashboard in place, (3) we have FlowFuse AI features that can take data that has been coming in and make predictions about what will occur at various points in time.

As I am conceiving this, it can be delivered as a Blueprint for customers who use Tables. A Node-RED instance can collect data from whatever source the customer would like to have predictive analytics for. The flow will push that data to Tables, and a Dashboard represents the historical pattern of that data. AI analyzes the data and provides predictions about future states. Another dashboard widget shows that future. An alerting mechanism can be configured if a future state is predicted to be too far away from the ideal state.

### Which customers would this be available to

Enterprise only