Skip to main content

Build Model

Moveo One makes it possible to build predictive models automatically — without writing code or managing training pipelines.

You simply define what you want to predict (the Target Event) and under what conditions (the Conditional Events), and the Moveo One backend handles data preparation, feature extraction, training, validation, and deployment.

Overview​

Each predictive model in Moveo One answers a question like:

GoalExample
Predict user conversion“Will the user complete checkout?”
Detect churn“Will this user return after 7 days?”
Identify confusion“Is the user likely to abandon onboarding?”

Models are powered by anonymized behavioral data captured from SDKs — event sequences, durations, transitions, and motion features.

Building via Dashboard​

Step 1. Open the Model Builder​

In the Moveo One Dashboard, go to:
AI → Predictive Models → Build New

Step 2. Choose a Target Event​

Select the main event you want to predict — for example:

  • checkout_complete
  • session_end
  • form_submit

Step 3. Add Conditional Events​

Pick 1–3 events that precede the target event, for example:

  • add_to_cart
  • view_product
  • checkout_start

These act as input signals for the model.

Step 4. Define Observation Window​

Set a time window (e.g., 30 minutes or 24 hours) that determines how far back in user activity the model looks for context.

Step 5. Click Build Model​

Moveo One automatically:

  • Selects relevant features
  • Balances datasets
  • Trains multiple algorithms (e.g., gradient boosting, logistic regression)
  • Chooses the best performer
  • Deploys the model for real-time prediction

Example Model Configuration (JSON)​

{
  "targetEvent": "checkout_complete",
  "conditionalEvents": ["add_to_cart", "checkout_start"],
  "observationWindow": "30m",
  "split": {
    "train": 0.8,
    "test": 0.2
  }
}

You can submit similar configurations via the API (see below).

Model Training Process​

  1. Feature Extraction — builds event sequences and statistical signals
  2. Training Phase — multiple models are evaluated and tuned
  3. Validation — best model selected based on accuracy/F1-score
  4. Deployment — endpoint automatically becomes active
  5. Monitoring — metrics appear in dashboard (latency, accuracy, drift)

Example Metrics Output​

{
    "standard_metrics": {
      "accuracy": 0.5501930501930502,
      "precision": 1.0,
      "recall": 0.20205479452054795,
      "f1": 0.33618233618233617,
      "auc": 0.9184901200145471
    },
    "optimal_threshold": 0.05
  }

Using the Model​

Once trained, predictions are available immediately vi SDK method: moveo.predict()

Retraining​

You can trigger a retrain manually if:

  • There has been a UX redesign
  • You’ve added new events or features
  • Model accuracy has degraded

Endpoint:

POST /models/retrain

Best Practices​

✅ Recommended

  • Ensure you have at least several hundred sessions per target event
  • Use balanced event frequencies to avoid bias
  • Choose observation windows that reflect real user flow (e.g., 10–60 minutes)
  • Retrain models monthly for fast-changing products

❌ Avoid

  • Using extremely rare events as targets
  • Building models before enough data has been collected

TODOs​

  • Add diagram showing model training pipeline
  • Add guide for scheduled retraining with CI/CD
  • Add link to dashboard “Build Model” walkthrough video