Overview

Analysis is the statistical and machine-learning capability inside the Analytics tab of the Chart Designer. Where Metrics overlay reference lines and conditional formatting on a chart’s visual surface, Analysis runs computations on the underlying data and surfaces the result as a summary, distribution, correlation, built model, or forecast. Analysis is the second capability under Analytics, alongside Metrics.

Analysis is accessed from Lens → Design → Chart Designer → Analytics → Analysis. Each analysis runs against a Datasource and dataset selected in the Navigator panel and dropped into the Designer’s Dimensions or Measure slot.

When to use it

  • Get a quick statistical overview of a dataset before deciding what to chart.
  • Identify outliers and distribution shape (box plot, histogram).
  • Quantify the relationship between two measures (correlation).
  • Group similar records automatically (clustering).
  • Predict a continuous value (regression) or a categorical value (classification).
  • Project future values from a time-series dataset.

Analysis types

SummaryDescriptive statistics for the dataset — counts, distributions, per-column statisticsNone.
Box PlotFive-number summary visualization (Minimum, Q1, Median, Q3, Maximum) showing data distribution.None.
PairwiseA pairwise comparison view.None.
CorrelationMeasure of the linear relationship between two data series.None — runs across the dataset’s measures.
HistogramFrequency distribution of values, grouped into user-defined bins.Number of Bins; Plot action.
ClusterGroups similar records into named clusters using a chosen algorithm.Model Name, Number of Clusters, Algorithm. See Build Model below.
RegressionPredicts a continuous value (the Prediction Column) using a chosen algorithm.Model Name, Select Prediction Column, Algorithm. See Build Model below.
ClassificationPredicts a categorical value (the Prediction Column) using a chosen algorithm.Model Name, Select Prediction Column, Algorithm. See Build Model below.
Time Series DecompositionTypically decomposes a time series into trend, seasonal, and residual components.None.
Time Series ForecastingProjects future values of a time series based on historical patterns.Forecast Length, Forecast Unit (Days / Weeks / Months / Year), Predict action.

Algorithm reference

K-Means ClusteringClusterPartitions data into a fixed number of clusters by minimizing within-cluster variance.
Hierarchical ClusteringClusterBuilds a nested hierarchy of clusters, useful when cluster count is uncertain.
Linear RegressionRegression, ClassificationFits a linear relationship between features and the prediction column.
Decision TreeRegression, ClassificationSplits the data into branches based on feature thresholds; produces an interpretable tree.
Random ForestRegression, ClassificationEnsemble of decision trees that votes on the prediction; generally more accurate than a single tree.
Support Vector MachineRegression, ClassificationFinds the boundary that best separates classes or fits the data, with optional non-linear kernels.
K-Nearest NeighborRegression, ClassificationPredicts based on the most similar records in the training data; non-parametric.
Naive BayesClassificationProbabilistic classifier based on Bayes’ theorem with strong independence assumptions between features.

Key behaviors

Different inputs from Metrics. Metrics overlay reference lines based on chart measures; Analysis operates on a full dataset placed in the Designer panel. The two capabilities can be combined on a single Chart — for example, a histogram from Analysis with an Average Line overlay from Metrics.

Algorithm choice affects interpretability and accuracy. Linear models (Linear Regression) are easier to interpret; ensemble models (Random Forest) are typically more accurate but harder to explain. The Algorithm field is per-model, so a Designer can build several models against the same data with different algorithms and compare results.

Time Series Forecasting depends on the unit matching the data. Selecting Months for daily data, or Days for monthly data, produces a forecast that doesn’t align with the natural cadence. Match the Forecast Unit to the dataset’s time granularity for meaningful projections.Correlation measures linear relationships only. A correlation value near zero does not mean the variables are unrelated — only that the linear relationship is weak. Non-linear relationships (quadratic, exponential, threshold) can still be strong despite low correlation values.