eazePredict API Documentation

Build prediction workflows for sales, demand, product usage, customer behavior, capacity, revenue, business metrics, operations, and service health.

The eazePredict API helps teams train models, run predictions, analyze time-series behavior, and support trust-aware forecasting workflows across the signals their business already creates: sales history, demand movement, product usage, customer actions, revenue patterns, capacity pressure, business metrics, and service-health signals when reliability is the workflow.

Authentication Copy cURL examples

Base URL

Prediction API base path

All code-verified JSON routes on this page use the v1 prediction prefix. Add the endpoint path after this prefix.

Base path 
/trace/engine/v1

Authentication

Use service-to-service Basic authorization

Protected routes under /trace/engine/v1/ require an Authorization header using the configured service token. The token value is deployment-managed through environment configuration and must not be exposed to browser clients.

Header 
Authorization: Basic <aiml-service-token>

Request format

Send JSON bodies with endpoint-specific request shapes

Training, prediction, and time-series routes use JSON request bodies. Include Content-Type: application/json, provide the endpoint-required model or series fields, and use timestamps and signal arrays in the documented request shape.

  • customerKey is required on train and saved-model prediction requests.
  • featureModel identifies supervised model artifacts for train and predict routes.
  • timeSeriesFeatureModel identifies time-series model artifacts.
  • externalFeatures and featureOutput describe inputs and target output for product, customer, sales, demand, operational, business, or service signals.
Required content type 
Content-Type: application/json
Example request body 
{
  "customerKey": "eazepredict-demo",
  "externalFeatures": [
    {
      "label": "website_visits",
      "output": [
        3000,
        3500
      ]
    },
    {
      "label": "discount_percent",
      "output": [
        15,
        20
      ]
    }
  ],
  "featureModel": "productSalesForecast",
  "featureOutput": {
    "label": "units_sold"
  },
  "startDate": "Wed, 01 Jan 2025 00:00:00 GMT",
  "timeSeriesFrequency": 1,
  "timeSeriesType": "d"
}

Response format

Responses are JSON and vary by route family

Successful routes return JSON with model results, forecast results, diagnostics, or error fields depending on the route. Training routes can include artifact metadata and trust summaries. Prediction routes return a model name and prediction rows. Time-series diagnostics return route-specific measurements such as suggested order, stationarity tests, decomposition parts, or forecast rows.

Example prediction response 
{
  "model": "linear",
  "prediction": [
    {
      "units_sold": 166,
      "timestamp": "Wed, 01 Jan 2025 10:00:00 GMT",
      "website_visits": 3000.0,
      "discount_percent": 15.0
    },
    {
      "units_sold": 181,
      "timestamp": "Wed, 01 Jan 2025 10:15:00 GMT",
      "website_visits": 3500.0,
      "discount_percent": 20.0
    }
  ]
}

Error handling

Handle common HTTP statuses without exposing internal traces

Error bodies are JSON. Treat validation and missing-artifact failures as application responses, and keep stack traces in server logs only.

Status Code Meaning Recommended action
200 / 201 Request completed successfully. Read the response body for model output, diagnostics, or created result details.
400 Invalid or incomplete request. Check required JSON fields, array lengths, content type, and timestamp formatting.
401 Missing or invalid authorization header. Verify the deployment-provided service token and header scheme.
404 Route or model artifact was not found. Check the endpoint path, model family, tenant key, feature model, and whether training has run.
500 Internal processing error. Retry if appropriate and inspect sanitized server logs using the request correlation id.

Endpoint group

Model Training

Training routes fit supervised regression models and save model artifacts for later prediction. Requests use the shared training shape with customerKey, featureModel, externalFeatures, featureOutput, startDate, and time-series cadence fields. The same pattern can support product usage, sales movement, demand, customer behavior, revenue, capacity, operational counts, business metrics, or service-health signals.

POST /train/model/linear

Train a Linear Regression model with feature arrays and a numeric target output.

Supports a simple baseline before using more expressive model families.

Open Linear guide
POST /train/model/polynomial

Train a Polynomial Regression model using the shared training request and optional degree.

Documented as a training route only. No code-verified prediction route is exposed for this family.

Open Polynomial guide
POST /train/model/poisson

Train a Poisson Regression model for count-style targets such as units sold, transaction counts, usage counts, capacity counts, operational totals, or service-health counts.

Targets should be non-negative counts for the best model fit.

Open Poisson guide
POST /train/model/decision_tree

Train a Decision Tree Regressor with optional tree hyperparameters.

Useful for threshold-driven behavior and non-linear business or operational patterns.

Open Decision Tree guide
POST /train/model/randomforest

Train a Random Forest Regressor with optional ensemble hyperparameters.

Useful when feature interactions are richer than a single model path.

Open Random Forest guide

Endpoint group

Prediction

Prediction routes load previously saved model artifacts and run inference on new feature arrays. Train the model family first for the same customerKey, featureModel, and optional subFeatureModel.

Method Path Description
POST /predict/linear Run inference with a saved Linear model.
POST /predict/decision_tree Run inference with a saved Decision Tree model.
POST /predict/randomforest Run inference with a saved Random Forest model.
POST /predict/poisson Run inference with a saved Poisson model.

Endpoint group

Time-Series Analysis

Time-series routes support forecasting, order selection, stationarity checks, random-walk baselines, and seasonal decomposition. These routes use signal arrays or structured time-series model requests for sales, demand, product usage, customer behavior, capacity, revenue, operations, and service health.

POST /timeseries/arima

Fit ARIMA and generate in-sample plus forecast rows.

Also supports /timeseries/arima/predict after a model artifact exists.

Open ARIMA guide
POST /timeseries/sarima

Fit SARIMA for seasonal time-series behavior and generate forecast rows.

Use /timeseries/sarima/predict for saved-model prediction.

Open SARIMA guide
POST /timeseries/autoarima

Automatically select a practical ARIMA order and fit the model.

Useful when order parameters are not known up front.

Open AutoARIMA guide
POST /timeseries/autoregression

Fit an AutoRegression model from lagged historical values.

Includes /timeseries/autoregression/predict for saved model inference.

Open AutoRegression guide

Endpoint group

Model Diagnostics and Forecasting Utilities

Diagnostic routes help select time-series parameters, inspect stationarity, decompose seasonality, and tune tree-based supervised models.

Method Path Use
POST /train/model/decision_tree/gridsearchcv Return best Decision Tree hyperparameters from GridSearchCV.
POST /train/model/randomforest/gridsearchcv Return best Random Forest hyperparameters from GridSearchCV.
POST /timeseries/arima/find_d Find a differencing order using ADF testing.
POST /timeseries/arima/find_p Estimate AR order from PACF analysis.
POST /timeseries/arima/find_q Estimate MA order from ACF analysis.
POST /timeseries/arima/findOrder Suggest a full (p,d,q) order and optional AIC grid result.
POST /timeseries/autoregression/find_p Estimate lag order for AutoRegression.
POST /timeseries/arma/find_p Estimate AR order for ARMA.
POST /timeseries/arma/find_q Estimate MA order for ARMA.
POST /timeseries/stationarity Run ADF and KPSS stationarity checks with a recommendation.
POST /timeseries/seasonal/decompose Return observed, trend, seasonal, residual, and seasonality-strength values.

Code examples

Copy-friendly cURL examples

These examples use verified endpoint paths and request fields. Replace the token, customer key, feature model, timestamps, and signal values with deployment-specific values.

Train a Linear model 
curl -X POST "/trace/engine/v1/train/model/linear" \
  -H "Authorization: Basic <aiml-service-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "customerKey": "eazepredict-demo",
    "featureModel": "weekly_product_sales",
    "externalFeatures": [
      { "label": "site_visits", "output": [1000, 1200, 1400, 1600, 1800] },
      { "label": "campaign_spend", "output": [420, 470, 530, 580, 640] }
    ],
    "featureOutput": { "label": "sales_units", "output": [22, 28, 31, 37, 43] },
    "startDate": "2026-04-01T00:00:00Z",
    "timeSeriesType": "h",
    "timeSeriesFrequency": 1,
    "train": 80
  }'
Predict with a saved Linear model 
curl -X POST "/trace/engine/v1/predict/linear" \
  -H "Authorization: Basic <aiml-service-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "customerKey": "eazepredict-demo",
    "featureModel": "weekly_product_sales",
    "externalFeatures": [
      { "label": "site_visits", "output": [1900, 2100] },
      { "label": "campaign_spend", "output": [660, 700] }
    ],
    "featureOutput": { "label": "sales_units" },
    "startDate": "2026-04-01T06:00:00Z",
    "timeSeriesType": "h",
    "timeSeriesFrequency": 1
  }'
Suggest ARIMA order 
curl -X POST "/trace/engine/v1/timeseries/arima/findOrder" \
  -H "Authorization: Basic <aiml-service-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "timeSeriesInput": [12, 13, 13, 14, 15, 15, 16, 19, 21, 24],
    "maxD": 2,
    "alpha": 0.05,
    "maxLag": 4,
    "gridSearch": true
  }'
Run stationarity diagnostics 
curl -X POST "/trace/engine/v1/timeseries/stationarity" \
  -H "Authorization: Basic <aiml-service-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "timeSeriesInput": [12, 13, 13, 14, 15, 15, 16],
    "significance": 0.05
  }'