Create a New ML View

The ML View feature in self-service visualization allows users to instantly generate Views based on machine learning tasks. By selecting the appropriate feature and target columns, users can apply ML algorithms directly to their datasets and visualize the results in reports.

Supported ML tasks include:

  • Segmentation (Clustering)

  • Anomaly Detection

  • Time Series Forecasting

  • Sentiment Analysis

This functionality enables business users to uncover patterns, detect outliers, predict trends, and analyze sentiment—all without leaving the visualization workspace.

Supported ML Tasks

Segmentation

  • Uses K-means clustering to identify hidden groups in data.

  • Input: Numerical features/measures.

  • Use case: Pinpoint customer segments, market groups, or product clusters.

  • Visualization Outputs:

    • Table: Compare up to two measures.

    • Scatter Plot: Explore relationships between two measures across clusters.

Anomaly Detection

  • Uses Isolation Forest or similar methods to detect outliers.

  • Input: Numerical features/measures.

  • Use case: Spot unusual data points (errors, deviations, or opportunities).

  • Visualization Outputs:

    • Line Chart: Highlight anomalies against overall trends.

    • Table: Show individual flagged rows for investigation.

Time Series Forecasting

  • Uses ARIMA/SARIMA (Auto-ARIMA) to predict future values.

  • Input:

    • 1 Date/Time column

    • 1 Measure column (e.g., sales, salary)

  • Requirement: Dates must be equally spaced (weekly, monthly, yearly).

  • Use case: Forecast sales, demand, or prices.

  • Visualization Outputs:

    • Table: Predicted values appended to the dataset.

    • Line Chart: Historical values with predicted points shown as dotted lines.

Sentiment Analysis

  • Uses BERT NLP models to classify text.

  • Input: Text/Dimension column.

  • Use case: Categorize sentiments in surveys, reviews, or social media (Positive/Negative).

  • Visualization Outputs:

    • Pie Chart or Mixed Chart: Overall sentiment distribution.

    • Tiles: Counts of positive and negative sentiments.

  • Note: Maximum supported word count = 2000.

Steps to Create a New ML View

  1. Navigate to the Report final screen.

  2. Click the Create New ML View icon.

  3. The Design View canvas opens.

  4. Select or drag-drop required Dimensions and Measures.

  5. Click GO to generate the View.

  6. Click the ML Properties icon (appears after the View is created).

  7. In the Machine Learning Properties window:

    • Select an ML Task from the menu.

    • Configure fields specific to the task (e.g., select feature/target columns).

    • For Segmentation:

      • Specify Cluster Count.

      • Adjust Cluster Colors using the color palette.

    • For Anomaly Detection/Time Series/Sentiment: Provide task-specific fields.

  8. Select the Operator from the context menu (where applicable).

  9. Click the Predict button.

  10. The ML-generated View is displayed.

  11. Click the Chart List icon to switch to a different chart type (e.g., scatter plot for Segmentation).

  12. Once satisfied, click Save View.

  13. A notification confirms the ML-based View has been added to the Report.

Visualization Results by Task

  • Segmentation: Scatter Plot or Table (grouped by clusters).

  • Anomaly Detection: Line Chart with anomalies highlighted, or a Table with flagged rows.

  • Time Series Forecasting: Line Chart (predicted vs historical) or Table (forecast values appended).

  • Sentiment Analysis: Pie/Mixed Charts with sentiment proportions; Tiles for positive vs negative counts.

Best Practices

  • Ensure date/time columns are equally spaced for time-series forecasting.

  • Limit text length for sentiment analysis to 2000 words.

  • Use scatter plots for segmentation to clearly distinguish clusters.

  • Always review outliers in anomaly detection to separate real issues from false positives.

Last updated