Skip to main content

​
What This Node Does

The Score ML Model node applies trained machine learning models to new data to generate predictions. Use evaluated models to score new customer records, predict prices, classify products, detect fraud, or forecast demand with probability scores and confidence intervals. [SCREENSHOT: Score ML Model node generating predictions on new data]

​
When to Use This Node

Use the Score ML Model node when you need to:
  • Make predictions on new data - Apply trained model to unlabeled records for production predictions
  • Batch scoring - Score thousands of customer records, products, or transactions at once
  • Real-time inference - Generate predictions for live data (new customer signup, incoming order)
  • Production deployment - Operationalize models for business use (dashboards, reports, alerts)

​
Step-by-Step Usage Guide

1

Add Score ML Model node

2

Connect evaluated model

Connect Eval ML Model node output to Score ML Model model input[SCREENSHOT: Eval ML Model connected to Score ML Model]
3

Connect new data

Add Input node with new data to score. Data must have same features as training data.[SCREENSHOT: Input data connected to Score ML Model]
4

Configure output

Classification: Set predicted column name, enable probabilities and confidence Regression: Set predicted value name, enable confidence intervals[SCREENSHOT: Output configuration]
5

Run and review predictions

​
Tips and Best Practices

Feature Consistency: Ensure scoring data has exact same features as training data. Missing or renamed features will cause errors.
Preprocessing Automatic: Score ML Model automatically applies same preprocessing (scaling, encoding) as training. Don’t preprocess manually.
Use Probabilities for Ranking: Don’t just use predicted class. Sort by probability to prioritize high-confidence predictions.
Monitor Prediction Confidence: Low confidence predictions (< 0.6) may need manual review or model retraining.
Batch for Throughput: Use batch scoring for large datasets. Real-time scoring for latency-sensitive applications only.
Version Control Models: Track which model version generated predictions. Retrain and rescore periodically as data changes.