Skip to main content

Modeling

Transform your time series data into powerful forecasting models. No coding required - just upload your data and let Heimdall Forecast automatically build the best model for your use case.

What You Can Forecast

  • Sales predictions - Forecast product demand and revenue
  • Resource planning - Predict server load, inventory needs
  • Financial forecasting - Predict stock prices, market trends
  • Weather patterns - Forecast temperature, precipitation
  • User behavior - Predict website traffic, app usage

How It Works

Heimdall's platform will guide you through uploading your time series data and selecting your target variable through an intuitive UI. Once you select your target, it will start building models that are best fit for your forecasting problem.

The entire forecasting pipeline is handled automatically:

  1. Data Preprocessing - Clean and prepare your time series data for training
  2. Feature Engineering - Extract relevant temporal features automatically
  3. Algorithm Selection - Test multiple forecasting algorithms to find the best fit
  4. Model Training - Train and optimize your forecasting model
  5. Validation - Test performance and ensure accuracy
  6. Deployment - Get a production-ready API endpoint

No coding or data science expertise required - just upload your data and get results!

Data Requirements

To build a forecasting model, you need to upload your time series data in the correct format.

File Requirements

  • File types: CSV or Excel (.xlsx) format
  • Headers: First row must contain column headers
  • Size: Minimum 200 rows, maximum 10 MB
  • Quality: Clean, structured time series data works best

Other Data Sources: For database connections and other data sources, see our Database Connections section.

Time Series Data Format

Your data should include both feature columns (the variables you'll use to make predictions) and a target column (the value you want to predict). Feature columns can be numerical, categorical, or text data. The target column should contain the actual values you want your model to learn to predict.

DateSalesMarketing_SpendSeasonTemperatureTarget_Sales
2024-01-0110005000Winter321200
2024-01-0212005500Winter351400
2024-01-0314006000Winter381600

This example shows a sales forecasting dataset where:

  • Date: Time series timestamp
  • Sales: Historical sales data
  • Marketing_Spend: Marketing investment
  • Season: Seasonal indicator
  • Temperature: External factor
  • Target_Sales: Target variable (what we want to predict)

Training

Models take different amounts of time to complete depending on your data size and complexity. Don't worry - you can come back anytime to check in on your model's progress. Once training is complete, you'll be prompted to save the model to your inventory. Heimdall will thoroughly analyze your time series data to find the best possible forecasting model for your problem.